22records
I Reduced Task Time
Reducing the time it takes an individual to perform a discrete task.
Benefits
Across the database: 600 cases, 600 with some field review, 1 with complete field review. Review coverage.
MapClose
Sector
Benefit
Beneficiary
Depth
Confidence
Claim
Harm on record
Yes: an incident database records the same kind of system harming the same kind of stakeholder, and the record links to it.
Country
Depth of benefit, by date
Depth12345depth unassessedBeneficiary a person an organization societyEvidence found or studied reported claimed
Dates are available for 466 of 600 cases.
Benefit type by depth
Benefit types
For a person
3records
II Reduced Task Cost
Reducing the financial costs to an individual to perform a discrete task.
3records
V Product Personalization
Customization or personalization of a product/service that increases usability or user value.
6records
VI Increased Creativity
Increased access to new ideas, or innovative ways of creating products/services.
118records
IX Improved Physical Health
Improved physical health outcomes or information about bodily health.
18records
XV Improved Learning
Improved educational outcomes or experiences including faster learning rates or academic achievements.
37records
XX Increased Product Safety
Improved safety features, detection mechanisms, or failsafes.
For an organization
5records
III Increased Scalability
Enabling a process to efficiently scale to additional users, sectors, or business units.
55records
IV Improved Task Accuracy
Increased accuracy or performance for a discrete task.
63records
VII Improved Knowledge Discovery
Easier and more accurate access to information necessary to accomplish a task.
1records
VIII Improved Customer Retention
Increased probability that a customer continues to use a product/service.
6records
XIII Repetitive Task Automation
Automating highly inefficient repetitive tasks, particularly those that contribute to harmful bottlenecks or bloated bureaucracy.
66records
XIV Reduced Risk
Reduction in potential for harm, especially potential legal harm.
58records
XVII Improved Organizational Efficiency
Improvements to overall operational efficiency, including helping facilitate cross team processes, reducing organizational friction, or reducing redundancies.
5records
XVIII Increased Product Revenue
Direct increases in product/service revenues or contract growth rates over time.
0records
XIX Improved Public Reputation
Improved social standing and reputation with core stakeholders.
For society
0records
X Increased Diversity
Improved diversity of communities or groups of people.
11records
XI Increased Inclusion
Increased acceptance of people with diverse backgrounds.
26records
XII Improved Environmental Sustainability
Improved environment including lowered carbon emissions, reduced waste, or using recycled materials.
46records
XVI Improved Accessibility
Increased accessibility in society for people with disabilities.
12records
XXI Reduced Carbon Emissions
Reduction in energy usage, water consumption, or direct carbon emission outputs.
5records
XXII Increased Access to Education
Improved access to educational services, institutions, or programs.
34records
XXIII Reduced Violence
Reduction in criminal activities or physically harmful incidents.
TableClose
Click a column head to sort the cases; click it again to reverse. The sort travels in the address.
CompareClose
Nothing selected yet. The Compare control beside each case adds it here. Selection cleared.
RolesClose
Read the cases the way your work needs them. A role adds its own line to every case and orders the list for it.
Latest by dateClose
2026
- Drivers slowed down when signs warned of wildlife aheadElk crossing: Real-time driver warning system helps prevent animal collisions, Published
- A father went to the ER after checking his heart with a smartwatchSmartwatch AFib alert sends dad to ER before heart surgery, Published
- Researchers found 47 known gravitational-wave signalsMADGRAV: a multilevel anomaly-detection pipeline for gravitational-wave searches applied to LIGO data, Published
- Nimbus's zasocitinib beats deucravacitinib in phase 3 trialZasocitinib demonstrates superior efficacy to deucravacitinib in LATITUDE Atlas, Published
- North East Derbyshire explores Copilot for routine workNorth East Derbyshire District Council: using Copilot to support better decisions and organisational change, Published
- Manchester residents shape a proposed AI advice serviceCitizens Advice Should Be the Trusted AI Advice Service, Published
- UNEP records eight methane mitigations in TurkmenistanMARS Mitigation Cases
- Caseworkers report easier reporting and case handlingWhat Team Moulton Taught Us About Building GovTech Right, Published
- Huddersfield's first facial recognition vans: three arrests, six checksCutting Edge Facial Recognition Cameras Used for First Time in Kirklees, Published
- Wearable rhythm strips let a 59-year-old go straight to ablationConsumer Wearable Device Leads to Expedited Diagnosis and Treatment of Paroxysmal Atrial Fibrillation, Published
- Transport for NSW trials PAWS wildlife cameras; no results yetNew AI animal detection and smart signage trial working to limit wildlife-vehicle strikes, Published
- o3-assisted reanalysis yields 11 new diagnoses and seven rediscoveriesLLM-Assisted Reanalysis of Unsolved Rare Disease Genomes Increases Diagnostic Yield, Published
- TIAA’s AI may have helped stop a scam involving retirement savingsAmerica has a $4 trillion retirement crisis and half of workers could run out of money. TIAA’s CEO wants ‘guaranteed lifetime income’ to fix it, Published
- DeterTech AI cameras may have cut trespass at DingwallAI cameras drive down Dingwall trespass but safety message remains, Published
Observed
Saronic Corsair sea drone picks up two Apache crew off OmanU.S. Army Crew Safely Rescued After Helicopter Lost at Sea, Published- Rapiscan CT algorithms detect marine wildlife at 86–96% in testingMarine wildlife trafficking: use of AI algorithms for the autodetection of shark fins, seahorses and sea cucumbers, Published
- AI planning may have helped SKMC separate craniopagus twinsPureHealth performs GCC's first separation of craniopagus twins' procedure in Abu Dhabi, Published
- ForeSight-HF scores two nursing-home heart-failure casesAI-Driven Heart Failure Decision Support in Skilled Nursing Facilities, Published
- Asilla's camera AI may have prompted two suicide interventionsJapan stations, facilities use AI system to prevent suicide by jumping, Published
Observed
OpenAI refers Darren Zhou's threats to the FBI; he pleads guiltyMan Flagged by ChatGPT Pleads Guilty in Stalking Case, Published- Apple Watch breathing alerts send a 79-year-old to CPAP therapySmartwatch-Detected Atrial Fibrillation on the Morning After a Nocturnal Sleep-Disordered Breathing Notification: A Case Report, Published
Information as of
AI generates property images for design discussionsStage, Style + Edit Real Estate Images with EDIT by DIY Render- WhaleSpotter thermal AI starts whale alerts in San Francisco BayNew AI system uses cameras and thermal sensors to steer ships clear of gray whales in the San Francisco Bay, Published
- Greece presents AI to transcribe and check GSM-R rail dialoguesΚυρανάκης στον ΣΚΑΪ: Σύστημα τεχνητής νοημοσύνης θα ελέγχει σε πραγματικό χρόνο τις συνομιλίες στα τρένα, Published
BriefsClose
- Ghost Murmur may have found downed F-15E officer
The April 6 White House briefing documents the rescue of a downed American airman in Iran.[1] The claim that an AI-assisted quantum sensor located him comes from two anonymous sources in the New York Post. President Trump subsequently told the Post that the tool was important, without explaining how it worked.[2] Physicists interviewed by the Post and Scientific American questioned the described capability.[3] The rescue is established; the contribution attributed to Ghost Murmur remains unverified.
ContributeClose
Readers can suggest a change to an entry and include the evidence that supports it.
Peter H. Vartanian reviews contributions before publication. Preparing a draft sends nothing.
How a record is madeClose
... which is why this database exists: to help others benefit from AI that has already helped someone else.
Edited by Peter H. Vartanian.
Admission
Cases concern reported or intended benefits with an identifiable AI contribution. These include AI-assisted learning, research, accessibility, and operational improvements. An AI-related topic alone is insufficient. Reported outcomes and intended benefits have distinct labels. Every case has a beneficiary shape, a depth color, and a confidence fill. Gray means that depth is unassessed.
Kinds of event
A record describes one of four kinds of case: a benefit created or restored, a specific harm prevented, a protective response whose prevention is unproven, or a general benefit case. General cases can include launches, simulations, accuracy studies, aggregate claims, cohorts, AI-assisted learning, and proposed applications. Their category does not determine their depth or confidence.
Depth
Each beneficiary group that benefited is rated from 1 to 5, the depth of the benefit for that group: 1, a convenience, or time or cost saved, with life otherwise unchanged; 2, a measurable improvement in a result or a capability, such as a test score or a crop yield; 3, a lasting change, a capability restored or a risk removed; 4, a serious harm averted or a serious condition resolved in time; 5, a life or a livelihood preserved, or a change documented across a population. Every rating has a reason. Depth describes the magnitude of the benefit that the source reports; it does not establish that the claim is independently verified. Aggregate outcomes and gains in capability can be rated. Intended benefits remain explicit, and a group stays gray when the extent of its benefit cannot be established. Corrections appear in the revision history.
The system's part
What the system contributed to the benefit is assessed separately from whether the benefit occurred, and may be left unrated. A system that was present is not thereby responsible.
A system’s name, base model, adapted variant, interface, hosting, and settings are assessed separately. A supported inference appears with its basis in an editor’s note. Later documentation may identify a plausible model, but does not establish that it was available or used on the event date. The importance of a task does not establish its reasoning settings.
Evidence and confidence
Source review follows citations, notes, appendices, and supplementary material to the study, dataset, document, or firsthand account that produced each finding. Different claims in one case may have different primary sources. An additional source must contribute evidence, verification, a correction, or a material limitation. Repetition does not constitute corroboration. A primary source may have an interest in the result, so origin and independence are assessed separately. When the original cannot be inspected, the record identifies the intermediary and the limitation.
Each source identifies its kind and its author's relationship to the system: maker, deployer, beneficiary, evaluator, independent author, or unassessed. An evaluator's role alone establishes neither independence nor a conflict of interest.
Only sources that support the benefit determine its evidence grade. Among independent sources, a court or regulatory finding gives the grade found. A peer-reviewed study, clinical case report, or research report gives studied. A descriptive report, news report, preprint, or dataset gives reported. The strongest qualifying source determines the grade. Without one of these sources, the grade is claimed. This includes accounts from interested parties, even when published as studies, and sources whose independence is unassessed.
The fill is solid for found and studied, hatched for reported, and outlined for claimed. These correspond to high, medium, and low confidence. Contrary evidence, procedural documents, and background material remain in the sources without raising the benefit's grade.
Source notes describe authorship, independence, and any documented conflicts of interest. They identify gaps in the cited evidence. Inferences are labeled.
Briefs
Briefs by Peter H. Vartanian examine the evidence and its gaps. Their dates appear in the revision history.
Replication
Each entry includes fields for the conditions that produced the benefit and the conditions that would have to hold again. Subsequent attempts appear with their setting, model version when reported, and outcome. An outcome that the cited sources do not establish remains unresolved.
Comparable incidents
Where an incident database holds an entry about the same kind of system harming the same kind of stakeholder, the record links to it, so that a reader weighing benefit against risk can open both. No severity is shown for the harm.
Exploring the cases
Cases can be filtered and viewed as a list, table, map, timeline, or table of benefit types. The view and the filters are kept in the address, so that a view can be shared.
Candidate reviews
The review log gives the reasons for excluding a candidate. Unfavorable results remain visible when they bear on a reported benefit. Multiple sources about the same benefit belong to one entry. Intended benefits, aggregate results, and weak evidence can qualify.
Dates
A case distinguishes an observed event date from the publication date, the date through which a source says that its information is current, or the date on which the source was checked. Month-only dates display the month without an invented day. The timeline uses that date only when an event or source supports it. Dates of access and revision do not determine the order.
Newest follows event or source dates. When a case's date basis has not been assessed, a dated source must support the same date before the case appears in Newest. That date is labeled as a source publication date. Source-check and revision dates do not determine the order. Cases without a supported date remain in the full database.
Coverage
Counts on this site describe the database's coverage. They do not measure how beneficial AI is, and a trend in records is a trend in reporting.
The map in the Gazetteer draws Natural Earth's 1:110m countries with de facto boundaries in an equirectangular projection. A country's circle has an area in proportion to its cases; a route runs from the country where a system was made to the country where the case happened. Cases that state no country are not on the map.
Not admitted
Excluded candidates and their reasons. An unresolved AI contribution can warrant further evidence. Intended benefits, aggregate results, and weak evidence can qualify.
Duplicate of a record
14
- Face2Gene, consumer wearables, speech neuroprostheses and the brain-spine interface AIBD-0037
These systems were not searched again because they are already in the database: Face2Gene (AIBD-0037), Apple Watch and other wearables (AIBD-0021 to AIBD-0027), speech neuroprostheses (AIBD-0002 to AIBD-0005) and the brain-spine interface (AIBD-0001). A new entry would repeat those records.
- OECD AI Monitor entry on Casey Harrell's UC Davis speech neuroprosthesis AIBD-0002
OECD AI Monitor entry 2026-06-15-4dfc covers the June 2026 press wave (Washington Post, 15 June; MIT Technology Review, 16 June) on Casey Harrell's UC Davis BrainGate2 speech neuroprosthesis, which is AIBD-0002, a record that already cites the NEJM 2024 paper and the Nature Medicine 2026 long-term-use paper.
- OECD AI Monitor entry on Meta's self-harm alert and Meerut police in eight minutes AIBD-0010
OECD AI Monitor entry 2026-06-08-7d81 (PTI via Rediff, ETV Bharat, The Hans India, LatestLY) describes the same event as AIBD-0010: Meta's alert to the UP Police Social Media Center at 6:53 pm on 7 June 2026 and Sardhana police at the 25-year-old's home within eight minutes.
- OECD AI Monitor entry on Mia flagging Barbara's 6 mm cancer at NHS Grampian AIBD-0062
OECD AI Monitor entry 2024-03-21-b742 (BBC, 21 March 2024; University of Aberdeen release) describes the same event as AIBD-0062, Mia flagging a 6 mm cancer missed by NHS Grampian radiologists in the GEMINI evaluation., 2026) and the 10 March 2026 Aberdeen release added to its evidence.
- OECD AI Monitor entry on TrailGuard AI and the 39 poachers arrested in Similipal AIBD-0102
OECD AI Monitor entry 2026-02-20-fae6 (The New Indian Express, Times of India and PTI, 20-21 February 2026) describes the same arrests as AIBD-0102; the agencies date the poachers' entry to Thursday night, 19 February, while the Field Director's account gives Wednesday night, 18 February.
- OECD AI Monitor entries on Indian Railways AI elephant detection AIBD-0009
Ten OECD AI Monitor entries on Indian Railways AI elephant detection (2023-03-20-b689, 2023-09-06-e510, 2023-11-29-0b17, 2024-02-09-5346, 2024-05-19-fc78, 2024-05-21-7e42, 2024-11-19-5a0a, 2025-08-12-50bf, 2025-12-20-724f, 2026-02-17-dcc6) repeat the Assam intrusion-detection case, AIBD-0009, and the Madukkarai thermal-camera case, AIBD-0101.
- OECD AI Monitor entry on Tesla Autopilot avoiding a wrong-way driver AIBD-0109
OECD AI Monitor entry 2020-03-07-752c, 'Tesla Autopilot Prevents and Fails to Prevent Collisions: Contrasting Incidents', corresponds to the Tesla Model S wrong-way-driver case in Cabarrus County already recorded as AIBD-0109.
- OECD AI Monitor entry on Tesla Autopilot braking as an oak fell in Storm Dennis AIBD-0108
OECD AI Monitor entry 2020-02-18-956e, 'Tesla Autopilot Prevents Fatalities During Storm Dennis by Emergency Braking', is the Tesla Model X oak-tree case already recorded as AIBD-0108.
- Tesla Autopilot Involved in Both Crash and Harm Avoidance Incidents (OECD Monitor 2020-06-08-c7f5)
The Monitor entry bundles the Taiwan overturned-truck crash (a harm; AIID 218) with the 'pig' avoidance video. The fetched Electrek (8 June 2020) and Daily Star (9 June 2020) articles show that video is Kenneth Van Hoesen's Belgian wild-boar swerve of 30 May 2020, which is written up from entry 2020-06-10-258f in this batch; no AIBD id exists yet for that record.
- AI System Enables Major Criminal Gang Bust in Chile's 'Operación Influencer' (OECD Monitor 2023-05-12-99fe) AIBD-0233
AIBD-0233 already records Operación Influencer (11 May 2023), the Fiscal Heredia model and the ten detentions, and cites this Monitor entry (2023-05-12-99fe) as a registry label.
- Tesla Autopilot Saves Driver During Heart Attack Emergency (OECD Monitor 2024-04-16-ba4d) AIBD-0139
Same drive as AIBD-0139 (registry id 2024-04-12-a558): Max Paul Franklin's Tesla Model Y on FSD v12 taking him to a VA emergency room on 2 April 2024. The Bosnian and Croatian reports behind this entry (Raport.ba, 16 April 2024, citing Elon Musk's 11 April post) retell that story as a 20 km heart-attack drive in Virginia; AIBD-0139 already notes the garbled syndication.
- Police use facial recognition AI to catch county lines drug dealers (OECD Monitor 2024-07-02-ee42) AIBD-0232
AIBD-0232 already records Cheshire Constabulary's facial recognition match to Keenan Hughes, the 17 May 2024 stop with Vincent Potter and the 1 July 2024 sentences, and cites this Monitor entry (2024-07-02-ee42) as a registry label.
- AI System in Japan Prevents Suicides at Train Stations and Commercial Buildings (OECD Monitor 2026-06-01-db16) AIBD-0234
AIBD-0234 already records Asilla's camera-analysis system at about 40 Japanese stations and buildings and the two suicide interventions reported by Kyodo on 1 June 2026, the same event and sources as this Monitor entry (Kyodo, Straits Times, Anadolu and syndications). AIBD-0234 carries no registry id, so 2026-06-01-db16 could be added to it.
- Paralyzed Man Controls Computer with Neuralink Chip, Spurs Safety and Privacy Worries (OECD Monitor 2025-03-23-f7ad) AIBD-0263
AIBD-0263 already records Noland Arbaugh's January 2024 Neuralink N1 implant and his cursor and chess control by thought, the same person, system and event as this Monitor entry; its malfunction and privacy material is the Monitor's ground for the incident label, not a separate benefit.
No AI system took part
9
- Spelling by a completely locked-in ALS patient (Chaudhary et al. 2022)
The Nature Communications paper states that a multichannel decoding algorithm was not implemented: letters were selected by firing-rate thresholds with auditory neurofeedback, and the only support-vector machine belonged to an earlier eye-tracking system. No machine-learning system produced the benefit.
- ChatGPT as manuscript editor for a missed-pneumothorax case report (Brown 2023)
ChatGPT was used only to draft and edit the Cureus paper about a delayed pneumothorax diagnosis at Jersey General Hospital, and the paper's AI content is a proposal of how AI could have helped. No AI system took part in the patient's care.
- BirdCast and the lights-out schemes of Raleigh and Bird Safe Philly
Raleigh's Lights Out (27 August 2026) follows a fixed seasonal schedule from 11 p.m. to 6 a.m., with BirdCast mentioned only as a public resource, and Bird Safe Philly's page neither mentions BirdCast alerts nor reports collision results, so no AI output changed either decision. The BirdCast-driven case, Houston's Lights Out Nights, is AIBD-0103.
- Titian catheter-robot ablation at Shaoxing People's Hospital (OECD AI Monitor entry)
The only event in OECD AI Monitor entry 2025-12-13-c088, the Titian robot ablation on a 64-year-old woman with paroxysmal atrial fibrillation at Shaoxing People's Hospital (Xinhua Zhejiang, 13 December 2025), used a teleoperated master-slave catheter robot steered by the surgeon at a console, and the article never mentions AI, autonomy or inference. The entry's CA-GPT item is a 96-patient comparative preprint rather than an event, and its third source concerns an unrelated neurosurgical robot in Fushun.
- Wandercraft Atalante exoskeleton walking for people with complete paraplegia
Independent studies exist, including the 2026 Journal of NeuroEngineering and Rehabilitation trial of 11 participants at Nimes University Hospital, but the exoskeleton walks on trajectories precomputed by nonlinear optimization and tracked by a hybrid zero dynamics controller, with no machine learning or AI involved.
- Estonia's small-claims 'robot judge'
The Estonian Ministry of Justice stated on 16 February 2022 that it does not develop an AI robot judge for small-claims or general court proceedings to replace the human judge, so there is no system and no outcome to verify.
- Rajesh Dandotiya bank-fraud intervention
The intervention in the Rajesh Dandotiya case was made by human bank staff rather than by an AI system, so there is nothing to record.
- Singapore Police Force Project A.S.T.R.O. scam alerts
The Singapore Police Force's own description (MHA Minister's Awards page, 29 October 2024) says Project ASTRO uses robotic process automation to send SMS alerts to scam victims, and the Straits Times (16 April 2023) describes officers and OCBC staff analyzing fund flows and mass-sending SMSes with it. The S$442 million said to have been averted since April 2023 is a police aggregate for automated messaging, not for model-based detection.
- AI-Enabled Halo System Prevents Serious Injury in Formula 1 Crash (OECD Monitor 2021-09-12-c119)
The Halo is a titanium cockpit bar that the FIA made compulsory from 2018; Wikipedia's entry and DER STANDARD's report of 13 September 2021 (Wolff: 'Der Halo hat definitiv Lewis' Leben gerettet') describe no sensor, software or AI component, and the Monitor's own sources, as listed, are match reports that call it a safety system. The 'AI-enabled' label is the Monitor's assertion alone.
Nothing to write up
14
- AI embryo selection or AI newborn screening changing an identifiable patient's care
No peer-reviewed case report of AI embryo selection or AI newborn screening changing an identifiable patient's care was found, so there was no system or case to examine.
- An individual patient helped by a sepsis early-warning alert
No individually documented sepsis early-warning case was found in a source independent of the developer; searches returned only cohort studies and implementation stories. The TREWS cohort evidence is recorded as AIBD-0066.
- ChatGPT run after the fact on a necrotizing otitis externa case (Pugliese 2023)
ChatGPT-3 and ChatGPT-4 were run on the Milan case only after the diagnosis had been made, did not suggest the correct diagnosis and had no part in the patient's care. The result is a failed test of the model, so there is no benefit to record.
- A Garmin watch alert leading to a diagnosis
No peer-reviewed case report of a Garmin alert leading to a diagnosis was found: Bogár et al. 2024 searched for Garmin and found none, and Europe PMC case-report searches returned no qualifying case.
- A WHOOP alert leading to a benefit
No peer-reviewed case report was found; only validation studies and vendor blog posts surfaced, with no case of a WHOOP alert leading to a benefit.
- AI-detected fast radio bursts at CHIME in 2024
Searches returned only machine-learning re-classification studies of the CHIME/FRB catalogs, not an AI-detected burst at CHIME in 2024 with independent confirmation, so there is no system or event to record. The Berkeley neural-network detection of FRB 121102 pulses is AIBD-0018.
- An AI painting attribution accepted by a museum
No museum statement accepting an AI-based attribution was located, and the candidate named no system, painting or museum; its other half, Ithaca, is AIBD-0017. The one named attribution case, Bradford's de Brécy Tondo, was examined under another theme and is AIBD-0276.
- AI identification of the language of the Voynich manuscript
No peer-reviewed acceptance of any AI identification of the Voynich manuscript's language was found, and no system, source or case was examined.
- Machine-learning attribution of a work to Bach
No peer-reviewed, scholar-accepted machine-learning attribution of a Bach work was found, and no system, source or case was examined.
- Sign-language recognition or translation deployed for deaf users
No deployed sign-language recognition or translation system with an independently evaluated outcome for deaf users was found. The 2026 JMIR systematic review of such systems for deaf health-care communication reports that every included system was at the development and testing stage and that contacted authors confirmed none had been deployed or commercialized.
- AI interception of a fraud or scam with a named victim's loss prevented
Searches returned only aggregate industry and regulator statistics on authorized push payment fraud; no court or regulator record of a specific AI interception with a named victim was found, and the candidate named no system, victim or event. The AI fraud-interception cases now in the database, such as TIAA's (AIBD-0210), came from other candidates.
- An individual automatic emergency braking intervention that prevented a crash
No regulator, court or peer-reviewed record of an individual AEB intervention in which the crash did not occur was found, and the NTSB documents located (Uber Tempe 2018, Ford BlueCruise 2026) record failures or disabled braking. The aggregate IIHS study is AIBD-0008; the Tesla incidents recorded from press reports (AIBD-0108, AIBD-0109) came from other candidates.
- UIC trial of AI trespass detection on railways
No UIC document naming an AI detection system or a trial result could be located: uic.org's safety pages list the TreSP network, TRESPAD events and project names such as AI4SAFEBEHAVE without any result, and its trespass-and-suicide page returned a 404 error. Nothing identifiable remains beyond the ProRail deployment, recorded separately as AIBD-0309.
- An evacuation or emergency decision taken on an AI flood or nowcast forecast
No authority statement documenting an evacuation or emergency decision taken on an AI flood or nowcast forecast was retrievable, and the candidate named no system, place or event of its own; the cases it pointed to are already AIBD-0031 (Bihar flood alerts) and AIBD-0092 (Mali skill evaluation). Actions on AI forecasts recorded from other candidates include the Kogi anticipatory cash payments (AIBD-0095) and the Philippine Red Cross activation (AIBD-0098).
A harm, not a benefit
1
- Netherlands SyRI welfare-fraud risk scoring
SyRI is a documented harm, a system the Hague District Court invalidated in 2020, and was set aside as such rather than examined as a benefit.
AI contribution to the benefit not established
1
- LGA-005 - Dorset council: the housing service digital transformation journey
The council's procurement assessment and Huume's service documentation connect the housing improvements to configurable workflows, policy calculations, portals, and dashboards. No reviewed source links an identified AI function to those gains. Dorset's separate meeting-summary and transcription pilots do not establish that AI contributed to this housing-transformation case.
DataClose
Data
The dataset is one file, records.json, released under CC BY 4.0. Three flat tables are derived from it: records.csv, one row per record; ratings.csv, one row per beneficiary group rated; and echoes.csv, one row per attempt to make a benefit happen again. The schema describes every field. Each record is also available on its own page as JSON.
Required coverage
Each case has 96 required field groups: 91 public and five internal editorial groups. Six collection-wide definitions apply through versioned collection profiles. Every field carries a value or an explicit assessment status. Unassessed questions remain visible and incomplete. Explore the fields and their meanings.
Download the public case profiles, field dictionary, and collection profiles. The coverage table lists each public field and its status for every case.
Fructus
AI outcomes, in money. A separately curated collection of financial claims, with original currencies, periods, recipients, comparisons, and supporting sources. JSON and CSV.
Counting
One system, one setting, one benefit is one record. A benefit reported in several sources is one record with several sources. A benefit that happened again elsewhere is an echo on the original record, not a new record.
Fields
- id
- A stable identifier.
- title, summary, date_observed, account
- What happened, in one line and in one paragraph, and when.
- brief
- A brief from Peter H. Vartanian: his reading of the evidence and its gaps, where one has been written, with its sidenotes in brief_notes.
- headline, metric
- A short title that adds no fact (asterisks around a name set it in italics), and the AI's quantified effect, copied verbatim from the summary or account (null when the record states no number).
- date_basis, date_precision
- Whether date_observed identifies an event, source publication, source information cutoff, or source access. Precision is day, month, year, or unassessed. A partial date uses the first day of its month or year only as a sorting anchor.
- benefit_status
- Reported, intended, mixed, or unclear. This is separate from evidence confidence.
- category
- The kind of event: benefit created, harm prevented, protective response (prevention unproven), or general benefit (including aggregate results and proposed applications).
- unexpected, purpose
- Whether the benefit was what the system was built to do, and what it was built to do.
- system, model_version, developer, deployer, location, domain, segment
- The system, its version where known, who made and ran it, and where it was used.
- benefit_type, taxonomy_note
- The benefit type from the taxonomy, with a note where the fit is loose.
- unrated_because
- For an event with no depth, the reason in a clause: the good is denied, or the AI's part is undocumented.
- ratings
- For each beneficiary group that benefited, a depth from 1 to 5 or null for unrated, and the reason.
- ai_role
- What the system contributed, assessed separately, or marked not assessed.
- evidence, confidence
- The sources include their kind, citation, link, date, a short quote, and author relationship: maker, deployer, beneficiary, evaluator, independent author, or a relationship that remains unassessed. Only sources that support the benefit determine confidence (found, studied, reported, or claimed). Shade (high, medium, or low) corresponds to a solid, hatched, or outlined fill. The method explains these grades. An evaluator's role alone establishes neither independence nor a conflict of interest.
- comparator
- Better than what: the baseline the benefit is measured against (against), the figure without and with the AI (before, after), whether the source measured or assumed the difference (basis), and the passage that states it (quote, url). Null when no comparison is recorded. The case profile distinguishes an unassessed question from a reviewed absence.
- later
- What happened after the first report: persisted, scaled, challenged, reversed, withdrawn, litigated or unknown, with the date, a note, and the passage that states it. A challenge disputes a claim without establishing that it was reversed. Unknown means that no later outcome is established in the record.
- value
- A financial benefit stated by a source: the amount verbatim with its currency, what it measures, who gains or pays, whether it is an estimate, and the supporting passage.
- claim_status, claim_basis
- The strongest source's authority, the cited evidence by kind and author, and its limitations. Both are computed from the evidence.
- replication
- The conditions that produced the benefit, the conditions that must hold again, and the echoes (replications): each attempt to repeat it, with setting, model version, and outcome.
- paired_incident
- Harm on record: an entry in an incident database about the same kind of system harming the same kind of stakeholder, with the database's name and a one-sentence basis for the link.
- source_collection, source_entry
- The imported collection and the stable entry key that AIBD assigns within it. These entry keys are not publisher-assigned identifiers.
- registry_ids
- Identifiers of the same event in other registries.
- revisions
- Every change, with its date, source, and editor.
Colophon
AI Benefits Database is made by Peter H. Vartanian. Write to peter.vartanian@icloud.com.
The dataset and this site are released under CC BY 4.0. Benefit types and beneficiary groups follow Trustible's benefits taxonomy.
The border adapts the color and grayscale reference strip shown beside a plate from E. A. Séguy’s Papillons in Smithsonian Libraries’ digitization.
Set in Newsreader, IBM Plex Sans and IBM Plex Mono, all used under the SIL Open Font License. Sector icons are Bootstrap Icons, under the MIT License. Organization logos are the organizations' own marks, shown to identify them; each one's source is listed in logos.json. Built 27 October 2026.
Cases
Full search could not load. Results below match titles only.
-
An Apple Watch called for help after a runner’s heart stopped
Category not assignedFall Detection on his Apple Watch Series 10 called emergency services, and he recovered fully after CPR and hospital treatment.
CPR at 6–7 min, full neurological recovery
-
An arthritis drug helped patients with severe COVID-19
Reported and intended benefitsFewer deaths
In RECOVERY, 12% of patients receiving baricitinib died within 28 days, compared with 14% receiving usual care alone. [2]
Faster recovery
In ACTT-2, patients receiving remdesivir recovered in a median of 7 days with baricitinib, compared with 8 days with a placebo. [5]
-
Lifeguards and a fellow swimmer rescued an unconscious man
Reported benefitLynxight sent his location and images from the pool’s cameras to a lifeguard’s smartwatch.
-
A man with a rare blood disorder became well enough for a transplant
Category not assignedMr. Joseph Coates received three drugs (i) that a consulting physician had considered and (ii) that Every Cure’s early drug-repurposing system had scored highly.
he responded within one or two weeks
-
Doctors found a nurse’s brain bleed with help from AI
Reported benefitJulie Knichel’s doctors received an alert after software, apparently Viz.ai, detected her brain bleed.
-
Ghost Murmur may have found downed F-15E officer
Category not assignedThe rescue is documented. The claim that Ghost Murmur located the airman rests on two anonymous sources and a subsequent presidential remark; its technical capability remains unverified.
one rescued airman; the claimed AI contribution remains unverified
-
Taiwan AI-ECG alerts reduce mortality in a randomized trial
Reported benefitA Taiwanese randomized trial involving 15,965 patients found 90-day mortality of 3.6% with AI-ECG alerts versus 4.3% with usual care.
3.6% versus 4.3%
-
COMPOSER sepsis alerts are linked to lower mortality
Reported benefitAt UC San Diego, mortality after COMPOSER deployment was 9.49%, versus a modeled 11.39% without it.
1.9 percentage points
-
AI helps locate Sophia Chan's aneurysm before emergency treatment
Reported benefitSophia Chan survived a ruptured aneurysm after AI-assisted diagnosis and emergency treatment at Mayo Clinic.
-
Smart wristwatch prompts ED readmission for pulmonary embolism
Category not assignedToronto Western Hospital orthopedic surgeons report a smart wristwatch prompting ED readmission for postoperative pulmonary embolism investigation and management; details are limited to the abstract.
-
Facial recognition helps HSI identify 311 probable child victims
Category not assignedA three-week HSI-led surge from 17 Jul 2023 produced 311 probable identifications of unknown child victims, 14 positive contacts and several rescues; officials credit facial recognition and AI images.
311 probable identifications of unknown child victims, 14 positive contacts
-
Vyntra's AI risk models may have stopped a USD 500,000 ATM fraud
Category not assignedVendor “Fraud Cases” page lists blocked payments from CHF 19,990 to USD 850,000 and a USD 500,000 (KSh 58 million) ATM fraud ending in arrests; customers, banks and dates are not given.
fraudsters were arrested while attempting to withdraw USD 500,000 (KSh 58 million)
-
WeChat Pay reminder stops a 190,000 yuan romance-scam transfer
Category not assignedWeChat says its anti-fraud reminder, with Tencent Cloud online service, led a Yunnan man to cancel a 190,000 yuan transfer to a “goddess” he had met online; reported by ChinaZ, 14 December 2021.
avoided a loss of 190,000 yuan
-
Spotlight may have led to a missing child's recovery in weeks
Category not assignedIn April 2019 a California officer ran a NCMEC missing-child photo through Thorn's Spotlight; Rekognition matching returned sex ads with the girl and, per Thorn's CEO, she was recovered within weeks.
claimed its tools helped identify 5,791 child victims and rescue 103
-
Meta self-harm alert gets Meerut police to a man in 8 minutes
Category not assigned7 June 2026: Meta alert at 6:53 pm; Meerut police reached a 25-year-old in 8 minutes, found him critical; hospital treatment, improved
3,011 lives saved statewide from 1 January 2023 to 31 May 2026
-
SARUAV finds an unconscious 65-year-old alive in tall grass
Category not assigned29 June 2021: SARUAV CNN flagged a 65-year-old man unconscious in tall grass after 24+ h; GOPR Bieszczady retrieved him alive
After 4 h 31 min of analysis the system detected him and returned his coordinates
-
ASENA flags a car at Yüksekova carrying 52.5 kg of explosives
Category not assignedPolice analysis software flagged a car at Yüksekova on 20 March 2022; it held 52.5 kg of explosives officials said were bound for Istanbul.
133 packets, 52.5 kg, of factory-made explosive
-
Meta flags a suicide post; Raebareli police arrive in 12 minutes
Category not assigned29 March 2026: Meta alert at 8:58 pm; Raebareli police reached an 18-year-old in 12 minutes after rat poison; hospital treatment, stable
A police team reached his home within 12 minutes
-
AI planning may have helped SKMC separate craniopagus twins
Category not assignedPureHealth/SKMC statement (5 June 2026) cites “AI-enhanced surgical planning”; XRlabs' own release claims AI-designed implants. The twins were separated in four operations and returned to Nigeria.
successful separation in four staged procedures
-
Medicare withholds $4.41 billion in suspected fraudulent payments
Reported benefitFederal analysts flagged unusual billing, and HHS-OIG and CMS held $4.41 billion in escrow during Operation Gold Rush.
HHS-OIG and CMS prevented all but $41 million of $4.45 billion
-
An early resistance warning from CMUH’s Intelligent Anti-Microbial System helped doctors change a 78-year-old woman’s antibiotic before she recovered.
flagged ceftazidime resistance in 1 hour
-
Kent reports fewer homeless households after targeted support
Reported benefitKent reports under 1% homelessness among proactively supported households versus 40% in an unsupported comparator. The groups were not randomized.
-
Mother's AI query prompts a rare-disease diagnosis in a 22-month-old
Category not assignedAn unnamed AI app suggested a skull abnormality, prompting genetic evaluation of a 22-month-old. The authors attribute the app to Med-Gemini-3D, but its model is unverified.
-
Smartwatch alert uncovers PAC cardiomyopathy; LVEF 15% to 50%
Category not assignedGöttingen: a smartwatch alert in an asymptomatic 34-year-old revealed frequent premature atrial contractions and LVEF 35%, later 15%; radiofrequency ablation restored LVEF to 50% at six months.
radiofrequency ablation restored LVEF to 50% at six months
-
Smartwatch ECG captures 165 bpm tachycardia; Emory diagnoses VT
Category not assignedA smartwatch ECG recorded a 165 bpm wide-complex tachycardia during palpitations; VT was confirmed on telemetry, an ICD placed and later shocked VT, and ablation stopped the VT (JACC Case Rep 2025).
recorded a 165 bpm wide-complex tachycardia during palpitations
-
Withings ScanWatch ECGs lead to a long-QT diagnosis and nadolol
Category not assignedWithings ScanWatch ECGs showing prolonged QT led to a KCNH2 long-QT diagnosis and nadolol for a father and daughter (Germany).
the watch displayed values such as QT 506 ms and QTc 457 ms
-
Apple Watch ECG T-wave leads a 76-year-old to quadruple bypass
Category not assignedKoster et al. (EHJ Case Reports, 30 Jan 2024): a patient-initiated lead-I watch ECG showed a negative T-wave normalizing after 22 min; exercise test, CT and angiography found 3-vessel disease; CABG.
he received four bypass grafts; recovery was uneventful
-
ALERTCalifornia AI may have helped contain the Trotter Fire
Category not assignedTrotter Fire, Bear Valley Springs near Tehachapi, 14–17 August 2023, 52 acres; the AI-detection claim and “nearly 4,000 acres” counterfactual come only from ALERTCalifornia's founder via Microsoft.
contain the Trotter Fire at 52 acres
-
Apple Watch SpO2 of 88% prompts CT that finds pulmonary embolism
Category not assignedApple Watch Series 7 SpO2 of 88% prompted CT angiography that found pulmonary embolism; anticoagulated and discharged (Japan).
Apple Watch Series 7 had shown an oxygen saturation of 88% in the early morning
-
Smartwatch alert leads to an ICD and finds a lung adenocarcinoma
Category not assignedA 58-year-old's smartwatch flagged an irregular rate over 180 bpm; NSVT was diagnosed from watch data, an ICD placed, and the post-procedure X-ray found a 28 mm lung adenocarcinoma, resected.
found a 28 mm lung adenocarcinoma, resected
-
Empatica Embrace seizure alert reaches Morgan McGrath's mother
Category not assignedIEEE Spectrum (6 February 2018): an Embrace wristband alerted the mother of Morgan McGrath, 26, of Fulton, Mo., during a seizure about 15 months earlier; she called paramedics and McGrath survived.
-
DeepSeek-R1 proposes FMF; MEFV testing confirms the diagnosis
Category not assignedPatient's DeepSeek-R1 dialogue proposed FMF; MEFV genetic testing confirmed it and colchicine stopped attacks after ~30 years of symptoms.
colchicine stopped attacks after ~30 years of symptoms
-
ML poverty maps select 1,163 wards for Nigeria's COVID-19 cash
Category not assignedML poverty maps chose 1,163 poorest urban wards for Nigeria's COVID-19 cash register; 49,000 beneficiary households as of 15 October 2021
49,000 beneficiary households of the government's COVID-CT as of 15 October 2021
-
STAR finds two motile sperm; couple conceives after 19 years
Category not assignedSTAR AI found 7 sperm (2 motile) in 2.5m images of a “sperm-free” sample; a couple infertile for 19 years conceived and had a girl.
found 7 sperm (2 motile) in 2.5m images of a “sperm-free” sample
-
A father went to the ER after checking his heart with a smartwatch
Category not assignedMr. Joseph Holycross reported that doctors confirmed the atrial fibrillation that Samsung Health Monitor detected on his Galaxy Watch4 Classic.
open-heart surgery followed ten days later
-
Saronic Corsair sea drone picks up two Apache crew off Oman
Reported benefitA Navy Corsair sea drone picked up two Apache crew off Oman and carried them to a helicopter hoist; CENTCOM reported that both were rescued within about two hours.
both were rescued within approximately two hours
-
Asilla's camera AI may have prompted two suicide interventions
Category not assignedAsilla told Kyodo that camera alerts prompted guards to intervene in two apparent suicide risks. The facilities and dates were not identified, and the company is the sole source of those accounts.
two reported interventions after alerts; dates and facilities undisclosed
-
OpenAI refers Darren Zhou's threats to the FBI; he pleads guilty
Category not assignedAn OpenAI disclosure brought Darren Zhou’s threats to police attention; he pleaded guilty to stalking and threat charges and received probation, GPS monitoring, and contact and weapons restrictions.
arrest on May 29, 2026; guilty plea on August 13
-
ChatGPT and Perplexity may have won Lynn White her eviction appeal
Category not assignedLong Beach tenant Lynn White used ChatGPT and Perplexity to overturn her eviction on appeal, avoiding about $55,000 in penalties, per NBC.
avoided roughly $55,000 in penalties and over $18,000 in overdue rent
-
ChatGPT names an insulinoma doctors missed for three years
Category not assignedChatGPT suggested an insulinoma after three undiagnosed years; a glucose sensor and ER blood tests confirmed it and surgery cured him.
surgeons removed about two thirds of his pancreas and his symptoms went
A muscle-signal “smart hand” returns daily tasks to Wang Bing
Category not assignedGuangming Daily (13 August 2024): Wang Bing, who lost both hands, uses the institute's algorithm-decoded “smart hand” to eat with a spoon, drink, shake hands, rotate his wrist and write his name.
Wexton addresses the House with an ElevenLabs clone of her voice
Category not assignedRep. Jennifer Wexton, who lost her speech to PSP, addressed the House on 25 July 2024 with an ElevenLabs clone of her own voice.
Neuralink N1 lets Noland Arbaugh control a cursor by thought
Category not assignedFirst PRIME-study participant, implanted at Barrow in January 2024, controls a computer by thought; threads retracted after a month, software fixes restored use; a second user drives a robot arm.
10 hours of daily use for chess, coursework, scheduling and communication at 28 months
AI colonoscopy flags a polyp; stage I cancer removed in Chiayi
Category not assignedAfter a positive fecal occult blood test, Fujifilm AI endoscopy at St. Martin De Porres Hospital, Chiayi, marked a 1 to 1.5 cm adenoma; pathology: stage I colorectal cancer, removed endoscopically.
a 1 to 1.5 cm adenoma that biopsy confirmed as stage I colorectal cancer
Mia flags a 6 mm cancer missed by NHS Grampian radiologists
Category not assignedIn NHS Grampian's GEMINI pilot, Mia flagged Barbara's 6 mm breast cancer missed by radiologists; surgery plus five days' radiotherapy.
Mia helped find 12% more cancers than routine practice
ML ink detection reads 2,000+ characters of a Herculaneum scroll
Category not assignedML ink detection let contestants read 2,000+ characters of a sealed Herculaneum scroll; papyrologists confirmed the text (Feb 2024).
recovered 15 columns and more than 2,000 characters of Greek text
CBP machine-learning flag at San Ysidro finds 75 kg of narcotics
Category not assignedCBP ML model flagged a car's crossing history at San Ysidro in 2023; review found 75 kg of narcotics and the driver was arrested.
review found 75 kg of narcotics and the driver was arrested
Bone-anchored Mia Hand cuts Karin's phantom pain from 5 to 3
Category not assignedSwedish amputee Karin uses a bone-anchored, nerve-linked Mia Hand daily; phantom pain fell 5 to 3, stump pain gone (Sci. Robotics 2023).
her phantom limb pain fell from 5 to 3, stump pain vanished
ChatGPT suggests tethered cord syndrome; neurosurgeon confirms
Category not assignedMother entered son's MRI notes and symptoms into ChatGPT; it suggested tethered cord syndrome, confirmed and operated on by a neurosurgeon.
Double neural bypass restores Keith Thomas's movement and touch
Category not assignedKeith Thomas, paralyzed since 2020, regained arm movement and touch after a 15-hour brain-implant operation in Manhasset, New York.
87% vs 27% successful grasp-force control with and without reinforcement learning
CMUH's exercise-ECG AI catches a 99% LAD stenosis rated moderate
Category not assignedCMUH's exercise-ECG AI put severe-blockage odds at 72% where manual reading said follow-up; angiography found a 99% LAD stenosis, stented.
angiography found a 99% LAD stenosis, stented
INEC's ABIS invalidates 2,780,756 ineligible voter registrations
Category not assignedINEC chairman Mahmood Yakubu reported on 26 Oct 2022 that ABIS cleaning of 2021–22 registrations identified 2,780,756 (22.6%) double, underage or fake entries, leaving 9,518,188 valid new registrants.
2,780,756 registrations (22.6%) were identified as ineligible and invalidated
DeepMind RL controller shapes TCV plasmas and sustains droplets
Category not assignedA deep RL controller commanding all 19 TCV coils produced and held elongated, negative-triangularity, snowflake and twin-droplet plasmas at EPFL's Swiss Plasma Center (Nature, 16 February 2022).
narrowed the separation to 6.6 cm with a combined RMSE of 3.7 cm
Mayo AI-ECG flags low EF; 41-year-old recovers to 55%
Category not assignedAt Mayo Clinic two ECGs scored 97.88% and 98.67% for reduced ejection fraction in a 41-year-old with no cardiac history; echo confirmed HFrEF, therapy followed, EF 55% at 4 months. Developer-authored.
at four months his ejection fraction had improved to 55%
Locust control fed by eLocust3m may have averted USD 1.56 billion
Category not assignedPlantVillage's app, built with FAO in under a month in 2020, gathered 240,000+ records; AI removed unusable data; FAO credits the tech-enabled operation with averting USD 1.56 billion in losses.
protected the food security of 36.6 million people and avoided USD 1.56 billion in losses
A brain–spine interface restored control of walking after paralysis
Reported benefitGert-Jan Oskam used the STIMO-BSI interface to stand, walk, and climb stairs after a spinal cord injury.
Walked more than 10 meters with crutches and no stimulation after rehabilitation.
Darpan reunites Som Soni with his family after five years
Category not assignedDarpan matched Som Soni, missing from Allahabad since 14 July 2015, to a boy in a Goalpara, Assam child home; reunited October 2020.
Two Tesla Model X cars stop as an oak falls; eight walk away
Category not assignedTwo Model X cars stopped as an oak fell on the A31 in Storm Dennis; eight occupants walked away, both drivers credit the emergency brakes.
eight occupants walked away
SkinVision flags a mole as high risk; stage 1A melanoma removed
Category not assignedSkinVision rated a 31-year-old's changing arm mole high risk; stage 1A melanoma was found and removed, and he is cancer-free.
stage 1A melanoma was found and removed, and he is cancer-free
Global Fishing Watch helps Kiribati fine a vessel US$1 million
Category not assignedKiribati used the Google-Oceana-SkyTruth platform to document Marshalls 203 fishing illegally in PIPA; a US$1 million fine plus a US$1 million goodwill grant followed (reported September 2016).
a US$1 million fine plus a US$1 million goodwill grant followed
Flint uses predictions to find hazardous water pipes
Reported benefitActiveRemediation helped Flint prioritize homes for inspection and replacement of lead or galvanized-steel service lines.
91.6% held-out prediction accuracy; 10.7% lower cost per replacement in simulation
UC Davis neuroprosthesis restores conversation for Casey Harrell
Category not assignedFour 64-electrode arrays plus RNN decoder let Casey Harrell (ALS) converse at 97.5% accuracy; later 1.96 million words at home
97.5% sustained accuracy over 248 hours of self-paced conversation
FRNSW drone AI finds two lost hikers in under five hours
Category not assigned23 June 2026: FRNSW drone AI flagged two lost hikers 500 m off track in Kosciuszko NP; found in under five hours, mild exposure
located less than five hours after the report
IBM–Yamagata AI survey confirms 303 new Nazca geoglyphs in Peru
Category not assignedAn IBM–Yamagata deep-learning survey guided fieldwork that confirmed 303 new Nazca geoglyphs, published in PNAS (Sep 2024).
303 previously unknown figurative relief-type geoglyphs, almost doubling the known total
PAWS-directed patrols remove over 1,000 snares in Cambodia
Category not assignedPAWS-directed ranger patrols in Srepok Wildlife Sanctuary removed over 1,000 snares in December 2018, versus 200–300 in a typical month.
removed over 1,000 snares in December 2018, versus 200–300 in a typical month
AI thermal cameras stop trains for elephants at Madukkarai
Category not assignedTN Forest Dept AI cameras on a 7.05 km rail stretch: 7,116+ alerts, thousands of train stops, zero elephant deaths since Feb 2024.
about 9,481 safe elephant crossings and no elephant deaths on the stretch
TrailGuard AI alerts lead to arrest of 39 poachers in Similipal
Category not assignedAI trail cameras detected armed intruders on 18 February 2026; STR Field Director reports 39 surrendered, nine guns seized, case registered.
at least 39 poachers surrendered
Apple Watch calls an ambulance for an unconscious motorcyclist
Category not assignedApple Watch Fall Detection called an ambulance and alerted his girlfriend after a hit-and-run left a Singapore motorcyclist unconscious.
Modus V Synaptive guides awake removal of a 4 cm brain clot
Category not assignedAwake surgery guided by the Modus V Synaptive robot removed a 58-year-old stroke patient's 4 cm clot at Tam Anh Hospital; full recovery.
About 40 ml of clot came out in just over 30 minutes
NEO BCI lets a man quadriplegic for 14 years drink unaided
Category not assigned54-year-old man, quadriplegic for 14 years, drank unaided by driving a pneumatic glove through the NEO epidural BCI; announced 29 January 2024.
drink water unaided, with grasp-decoding accuracy above 90%
NIO's NWM emergency pull-over stops a sleeping driver's car
Category not assignedDriver asleep at 110 km/h on a rainy highway; NIO's NWM emergency pull-over stopped the car and an SOS call woke him, 1 June 2025.
Daejeon's Kkumdori robot hears “I want to die” and brings police
Category not assignedAt 2 a.m. on 6 August 2025 Daejeon's Kkumdori care robot flagged a 70s woman's “I want to die”; police came and she was admitted next day.
Grok tells a man sent home to demand a CT; appendix near rupture
Category not assignedGrok told a 49-year-old Norwegian sent home with an acid blocker to demand a CT; it found an appendix near rupture, removed 3 December 2025.
Frankfurt camera AI finds a missing 16-year-old within a day
Category not assignedFrankfurt police found a missing 16-year-old in crisis within a day via a court-backed AI face search of Bahnhofsviertel cameras.
found a missing 16-year-old in crisis within a day
thynC flags ventricular tachycardia; CPR restores circulation
Category not assignedthynC alarm flagged ventricular tachycardia in an 85-year-old on a Gwangju general ward; staff CPR restored circulation, 14 January 2026.
a heart rate that climbed to 192 beats per minute
Tesla FSD may have pulled over after Rishi Vohra passed out
Category not assignedRishi Vohra says FSD pulled his Cybertruck onto a freeway shoulder after he passed out, 18 February 2026; Musk replied, no data published.
IMIE intelligent retina restores partial vision to a blind woman
Category not assignedBlind over a decade, a 61-year-old woman read E optotypes and walked indoors unaided after a 256-channel IMIE retinal implant in Changsha.
she could identify E optotypes (acuity 0.03, peak 0.1)
Fiscal Heredia may have identified a 10-member robbery gang
Category not assignedIn Operación Influencer (11 May 2023) prosecutors and PDI detained a gang of 10 aged 15–22 behind eight accredited violent robberies; the Fiscalía credits the university-built AI model Fiscal Heredia.
detained a gang of 10 aged 15–22 behind eight accredited violent robberies
OpenAI flag leads Brazilian police to hold a father a day early
Category not assignedOpenAI flagged a user's plan to kill his 8-year-old son; relayed via the FBI, Brazilian police held him a day before his stated date.
held him a day before his stated date
Apple Watch Crash Detection calls 911 for Andi Burns in a river
Category not assignedOn 22 February 2026 near Batavia, Ohio, a 19-year-old's truck flipped into the Little Miami River; Clermont County officials say her Apple Watch called 911 by Crash Detection and crews freed her.
Apple Watch Fall Detection calls 911 for hiker James Prudenciano
Category not assignedOn 14 October 2019 a lost hiker fell about 30 ft in Hartshorne Woods Park, NJ; his new Apple Watch's Fall Detection called 911 and texted his family; Middletown police say three agencies responded.
Nimbus's zasocitinib beats deucravacitinib in phase 3 trial
Category not assignedTakeda reports 36.5% complete skin clearance with zasocitinib versus 13.9% with deucravacitinib at week 16. The contribution of AI to discovery is not separately measured.
PASI 100 36.5% vs 13.9% for deucravacitinib at week 16
TIAA’s AI may have helped stop a scam involving retirement savings
Category not assignedTIAA’s CEO said that an AI flag, staff intervention, and contact with a participant’s daughter protected retirement savings of more than $3 million.
retirement savings above $3 million retained, according to TIAA’s CEO
EchoNext ECG flag finds severe aortic stenosis, leads to TAVR
Category not assignedAt Columbia/NewYork-Presbyterian, EchoNext flagged a pre-operative ECG (score 0.6094), triggering an EHR advisory; echo showed LVEF 40% and severe aortic stenosis; cardiac anesthesia, surgery, TAVR.
echo showed LVEF 40% and severe aortic stenosis
NHC blends DeepMind's ensemble into Hurricane Melissa forecasts
Category not assignedNHC blended Google DeepMind's AI ensemble into official Melissa forecasts from its first advisory on 21 October 2025; Cat 5 landfall in Jamaica 28 October 2025.
AI lip-reading training reports restored speech
Reported benefitHuazhong University reports that 343 deaf users regained speech capabilities with AI-guided lip-reading practice. The university also claims threefold learning efficiency.
Viz ICH Plus flags a haematoma; transfer in 101 minutes, mRS 1
Category not assignedViz ICH Plus flagged a 47-year-old's haematoma and alerted the transfer team; transfer in 101 min vs 199.7 average; mRS 1 at one year.
transfer in 101 min vs 199.7 average; mRS 1 at one year
Viz HCM flags hypertrophic cardiomyopathy in a 61-year-old
Category not assignedAI-ECG (Viz HCM) scored >0.95 for HCM in a symptomatic 61-year-old; echo confirmed obstructive HCM; mavacamten cut gradients and symptoms.
by week 8 the resting gradient was 9 mm Hg
SpliceAI nominates a deep intronic PKHD1 variant in fatal ARPKD
Category not assignedSpliceAI scored a deep intronic PKHD1 variant 0.98; RNA studies confirmed aberrant splicing; parents used the diagnosis for IVF with PGT-M.
a maternally inherited deep intronic PKHD1 variant scoring 0.98
RapidAI LVO alert gets Eric's artery reopened within 60 minutes
Category not assignedRapidAI's automated LVO alert mobilized the stroke team for Eric, 29; thrombectomy restored flow under 60 minutes from arrival; no deficits.
thrombectomy restored flow under 60 minutes from arrival; no deficits
MiniMed 640G cuts hypoglycemia; mother delivers a 4,137 g boy
Category not assignedMiniMed 640G predictive low-glucose suspend cut hypoglycemia in a T1D woman after two miscarriages; term birth of a 4,137 g boy.
hypoglycemic episodes fell from 2.1–2.5 to 1.4 per day
People with severe paralysis regained control of a computer at home
Reported benefitTwo people with ALS regained control of a computer at home using Stentrode and eye tracking.
typed 13.81 and 20.10 correct characters per minute
An implanted interface restored communication for nearly seven years
Reported benefitA woman with locked-in ALS used the Utrecht NeuroProsthesis to communicate at home for nearly seven years.
seven years of independent at-home use
Fabric GEM flags ANTXR2 and diagnoses an infant in five days
Category not assignedAI-assisted rapid genome sequencing scored a homozygous ANTXR2 variant at 4.7, diagnosing an infant within five days and changing her care plan.
scored a homozygous ANTXR2 variant at 4.7, diagnosing an infant within five days
Lunit INSIGHT CXR finds 13 resectable lung cancers incidentally
Category not assignedAI on routine chest X-rays flagged nodules in 13 patients not suspected of lung cancer; all had resectable cancer, 7 stage I, and surgery.
13 (17.3%) had cancers found incidentally because the software flagged a nodule
Brainomix flags Jean Hines's stroke; transferred in 25 minutes
Category not assignedAI-supported scan identified Jean Hines's major stroke on arrival; transferred in 25 minutes for successful thrombectomy; good recovery.
transferred in 25 minutes for successful thrombectomy
Imagene AI infers EGFR from H&E; osimertinib 2 days post-biopsy
Category not assignedImagene AI predicted EGFR from an H&E slide in <48 h; rapid PCR confirmed; osimertinib began 2 days post-biopsy, no neuraxis radiotherapy.
one week earlier than routine practice and avoided whole-neuraxis radiotherapy
qXR nodule flags lead to five early-stage lung cancer diagnoses
Category not assignedqXR flagged nodules on X-rays done for other reasons; CT/biopsy confirmed early-stage lung cancer in five adults; four had surgery.
CT/biopsy confirmed early-stage lung cancer in five adults; four had surgery
UNEP MARS alert leads to methane leak repair at Hassi Messaoud
Category not assignedMARS alert led an Algerian operator to fix an unlit flare at Hassi Messaoud; satellites show no emissions since 15 October 2024.
Novissi targeting pays 138,531 rural beneficiaries in Togo
Category not assignedML on satellite imagery and phone metadata set eligibility for Novissi rural cash; 138,531 beneficiaries paid, Nov 2020 to Aug 2021
reduced exclusion errors by 4–21% relative to geographic alternatives
Wearable rhythm strips let a 59-year-old go straight to ablation
Category not assignedVirginia family physicians report that AF episodes a 59-year-old recorded on a consumer wearable, confirmed by a cardiologist, expedited referral and ablation; AF-free off drugs at follow-up.
ChatGPT ties ten-plus years of symptoms to an MTHFR mutation
Category not assignedA Reddit user says ChatGPT tied 10+ years of symptoms to an MTHFR A1298C result and B12 use; doctor agreed; symptoms largely resolved.
Smartwatch AF alerts in a 25-year-old lead to cardioversion
Category not assignedA 25-year-old's smartwatch notified him several times of an irregular rhythm consistent with AF; ECG showed AF with rapid ventricular response, treated by TEE-guided cardioversion (Cureus, 2023).
atrial fibrillation with rapid ventricular response at 123 bpm
Fitbit Alta HR heart-rate rise leads to vasculitis diagnosis
Category not assignedFitbit Alta HR resting-heart-rate rise from 55 to 75 bpm led to early diagnosis and treatment of large vessel vasculitis (Zurich).
resting heart rate had risen persistently from about 55 to about 75 beats per minute
Smartwatch alert under 40 bpm reveals ciprofloxacin bradycardia
Category not assignedA smartwatch alert for a heart rate under 40 bpm two days after discharge on ciprofloxacin led to an ED visit; telemetry showed 30–50 bpm, the antibiotic was switched and the rate normalized.
by the next day her heart rate was 60–70 bpm
Fitbit Sense flags a 164 bpm rhythm, leading to a VT diagnosis
Category not assignedFitbit Sense recorded an abnormal rhythm before clinicians diagnosed and treated ventricular tachycardia in India in October 2021.
detected an irregular heart rate of 164 beats per minute with irregular rhythm
Apple Watch alarms lead to an AF diagnosis and anticoagulation
Category not assignedApple Watch heart-rate alarms led to an ECG-confirmed AF diagnosis and anticoagulation (UK, 2019); 16 echoes from 15 published case reports to 2026.
her pulse exceeded 120 beats per minute while she was inactive
Delhi students improve their math and Hindi scores
Reported benefitStudents offered free access to Mindspark centers gained 0.37 standard deviations in math and 0.23 in Hindi over 4.5 months.
+0.37 SD math, +0.23 SD Hindi in 4.5 months
More incoming students complete college enrollment
Reported benefitAmong students already committed to Georgia State, Pounce increased enrollment by 3.3 percentage points.
Enrollment rose by 3.3 percentage points among students already committed to attend
Higher algebra scores in the trial’s second year
Reported and intended benefitsStudents using Cognitive Tutor Algebra I scored 0.21 standard deviations higher in the second year, with no statistically significant difference after correction for multiple comparisons.
Second-year algebra gain: 0.21 standard deviations; inconclusive after correction
AI intrusion alert stops Kamrup Express as about 60 elephants cross
Category not assigned16 October 2024: AI intrusion alert and crew braking stopped 15959 Kamrup Express as about 60 elephants crossed near Lamsakhang, Assam
saved 414 elephants in 2023 and 383 up to 16 October 2024
Galaxy Fit pulse of 188 bpm leads to WPW diagnosis and ablation
Category not assignedGalaxy Fit recorded a resting pulse of 188 bpm; a routine ECG next day showed WPW and the band's records supported referral for successful ablation (South Korea, 2020).
recorded a peak resting pulse rate of 188 beats per minute
o3-assisted reanalysis yields 11 new diagnoses and seven rediscoveries
Category not assignedAI-assisted reanalysis of 376 locally unresolved cases yielded 18 confirmed diagnoses. Eleven were new, while seven had already been established elsewhere.
gave 18 families confirmed rare-disease diagnoses
Origin Orbit trial may have found over 800 Thames Water leaks
Category not assigned18-week Origin Orbit trial on Thames Water's network: over 800 leaks found, est. 8.7 million liters a day saved (utility figures); 13-month contract followed, target 100 million liters a day.
over 800 leaks across the network, saving an estimated 8.7 million liters of water a day
UNEP MARS alerts lead operators to divert gas from Iraqi flare
Category not assignedThe Chemical Engineer (5 May 2026), relaying UNEP: the largest source mitigated in 2026 was an Iraqi flare, 1.7m tCO2e since 2023, apparently in the Zubair field; Eni did not confirm closure.
the flare had released an estimated 1.7m tCO2e since 2023
APS AI camera spots Diamond Fire; crews hold it under 7 acres
Category not assignedAP: an APS AI camera detected smoke on a March 2026 afternoon and crews contained the Diamond Fire before 7 acres; the Arizona Daily Sun's account of 26 March 2026 gives the timeline but no camera.
contained it before it grew past 7 acres (2.8 hectares)
BrainGate2 neuroprosthesis lets a paralyzed man type 22 wpm
Category not assignedTwo BrainGate2 participants typed by attempted finger movement at up to 22 wpm, 1.6% word error, in at-home research sessions (Mar 2026).
T18 reached 110 characters (22 words) per minute with a 1.6% word error rate
ChatGPT and AlphaFold pick vaccine targets; dog's tumor halves
Category not assignedAI-chosen targets, a UNSW-made mRNA vaccine and a checkpoint inhibitor halved Rosie's largest tumor by early 2026; partial, not a cure.
halved Rosie's largest tumor by early 2026
Gyeonggi AI pig-weighing pilot lifts income per head 18%
Category not assigned2025 provincial pilot on six farms in Anseong, Pocheon and Yeoju: AI non-contact weighing cut shipping age 182.3 to 176.7 days, lifted 1+ grade share 35.3 to 37.7% and income per head 18%.
income per head rose from 100,000 to 118,000 won (18%)
LEADERG's AI door god flags smoke; temple fire held to 10 m²
Category not assignedAI camera analytics flagged smoke in an empty temple office before a battery fire; held to 10 m², no injuries, main hall intact.
held to 10 m², no injuries, main hall intact
ChatGPT flags fake adviser; victim reports Rs 67.10 lakh fraud
Category not assignedAfter withdrawals were refused, a 77-year-old Binnypet resident checked the firm's credentials with ChatGPT, which flagged complaints and inconsistencies; he reported the Rs 67.10 lakh fraud to Ben...
curl merges about 50 bugfixes from ZeroPath-assisted scans
Category not assignedIn Sept 2025 curl received hundreds of candidate issues found mainly with the ZeroPath AI analyzer by Joshua Rogers; maintainer Daniel Stenberg merged about 50 bugfixes from the first list.
about 50 separately identifiable bugfixes were merged
Veesion alert may have led police to a Szeged shoplifter
Category not assignedAI cameras in Szeged's unstaffed-at-night Coop flagged a shopper hiding goods; police caught them within hours, the operators said.
over four and a half months the system had exposed three theft attempts
ChatGPT sends a woman with a spider bite to the ER; staff agree
Category not assignedChatGPT sent Holli of Wolfforth, Texas to the ER for a worsening spider bite on 27 August 2025; doctors agreed and said the tissue was dying.
ADIAT spots a red helmet and locates Nicola Ivaldo on Monviso
Category not assignedAI analysis of 2,600 drone frames flagged a red helmet in the Perotti couloir at 3,150 m; Ivaldo's body was recovered by 31 July 2025.
located at about 3,150 m and recovered by 31 July 2025
ChatGPT traces an 18-month cough to a BP drug; doctor confirms
Category not assignedChatGPT flagged a BP drug side effect behind an 18-month cough; a doctor confirmed it and changed the drug; daughter reports mother healing.
Big Sleep may have isolated SQLite zero-day CVE-2025–6965
Category not assignedGoogle says Big Sleep, fed indicators that attackers were staging a zero-day, isolated SQLite CVE-2025–6965; SQLite shipped the fix in 3.50.2 on 28 June 2025, announced 15 July 2025.
Police phone app identifies man wanted since 2012 in Cardiff
Category not assignedA South Wales Police officer's OIFR phone app matched a 38-year-old in Cardiff to a 2012 Polish European arrest warrant for drug dealing and assaulting an officer; he was remanded (July 2025).
OpenAI o3 finds CVE-2025–37899, a use-after-free in ksmbd
Category not assignedSean Heelan, benchmarking o3 on ksmbd with no scaffolding, had the model surface an unknown use-after-free in the SMB logoff handler; the kernel CNA published CVE-2025–37899 on 20 May 2025.
KEA's facial recognition app exposes a forged CET admit card
Category not assignedKEA's facial recognition app flagged a woman with a forged CET admit card in Bengaluru on 17 April 2025; she fled before writing the paper.
Delhi Police facial recognition leads to Rs 79.5 lakh recovery
Category not assignedFacial recognition on CCTV named a Chandni Chowk robber; two arrested within about a day and Rs 79.5 lakh of the Rs 80 lakh recovered.
two arrested within about a day and Rs 79.5 lakh of the Rs 80 lakh recovered
ASENA flags a car at a Şırnak checkpoint; 86.6 kg of skunk found
Category not assignedŞırnak police stopped a car selected by the ASENA narcotics analysis software; the dog Hector found 86.6 kg of skunk in 119 packages in the trunk, and three lawyers were remanded (March 2025).
86.6 kg of skunk in 119 packages
AI-drafted objection gets a Tokat bus driver's traffic fine annulled
Category not assignedA Tokat school bus driver used ChatGPT to write his objection to a 2,400 TL fine for dropping a student off outside a stop; the court annulled it within three weeks for lack of evidence.
CGU's ALICE flags R$19.4m overpricing risk in three Rio tenders
Category not assignedCGU ALICE alerts: three 2024 federal hospital tenders in Rio de Janeiro state suspended, revoked or re-estimated; R$19.4m overpricing risk flagged (sum of three confirmed risks)
a saving of R$3,063,774.84
Grok spots a displaced wrist fracture that urgent care missed
Category not assignedBy her mother's X post of 11 January 2025, Grok flagged a displaced wrist fracture urgent care missed; an orthopaedist confirmed and cast it.
CGU's ALICE alert leads to a R$100 million cut in a road-works tender
Category not assignedA CGU ALICE alert on a federal road-works tender led managers to cut its reference value by about R$100 million, per CGU's 30 December 2024 results release
reduced the reference value by about R$100 million
Code Intelligence's Spark finds a heap use-after-free in wolfSSL
Category not assignedSpark, an AI fuzz-test agent, found a heap use-after-free in wolfSSL in late Oct 2024; fixed in PR 8139 and shipped in 5.7.6 on 31 December 2024.
Pano AI alert keeps Wellington fire to one-eighth of an acre
Category not assignedXcel-funded Pano AI camera alerted crews to a campsite fire east of a Breckenridge neighborhood; firefighters said the early notice kept it to one-eighth of an acre (CBS Colorado, 9 July 2024).
the early notice kept it to one-eighth of an acre
DeepMind's FunSearch finds a size-512 cap set in dimension 8
Category not assignedFunSearch produced a size-512 cap set in dimension 8 and new asymptotic bounds, published in Nature with independent commentary.
cap set of size 512 in dimension 8, exceeding the previous best known construction of 496
Shenzhen students' goose AI may have lifted survival 30%
Category not assignedSixteen Shenzhen University students with Tencent engineers built a camera system from 6,000 labeled images; Xinhua (13 Oct 2023) reports real-time alerts and 30% higher survival at the cooperative.
helped the farm increase survival of lion-head geese by 30 percent
HelioLinc3D finds asteroid 2022 SF289 in sparse ATLAS detections
Category not assignedHelioLinc3D linked sparse ATLAS detections into PHA 2022 SF289, published by the Minor Planet Center in MPEC 2023-O26 (Jul 2023).
about 600 feet across with an Earth MOID of 0.0015 AU
GPT-4 lists IMHA among causes of a dog's anemia; vet confirms
Category not assignedGPT-4 listed IMHA among causes of a dog's worsening anemia; a second vet confirmed and treated; the dog recovered, per the owner's thread.
UN INSARAG teams in Adıyaman may have used xView2 damage maps
Category not assignedMIT Technology Review (20 February 2023), quoting DIU: xView2 used by two UN INSARAG teams in Adıyaman and by AFAD, World Bank, IFRC and WFP; Microsoft reports 3,849 damaged buildings mapped for AFAD.
estimating 3,849 buildings damaged or destroyed
Palantir-based DAR identifies the author of a Siegen amok threat
Category not assignedNRW Interior Ministry, 28 Oct 2022: Palantir-based DAR resolved an amok-threat email against Siegen's city administration in seconds and identified 51 of about 100 people at a Duisburg clan shooting.
51 of the roughly 100 participants
AlphaTensor finds a 47-multiplication 4x4 scheme over GF(2)
Category not assignedAlphaTensor found a 47-multiplication 4x4 scheme over GF(2) and a 96 for 5x5; Kauers and Moosbauer improved the latter to 95.
multiplying two 4x4 matrices over the field with two elements in 47 multiplications
N2N face search finds Geddy Lee's mother in Bergen-Belsen photo
Category not assignedN2N face search found Geddy Lee's mother in a Bergen-Belsen DP-camp photo; Lee confirmed it and found more family photos at Yad Vashem
Google DR screening in Thailand matches specialists at 94.7%
Category not assignedAt nine primary-care sites (Dec 2018-Mar 2020) Google's deep-learning system graded fundus photos in real time, 94.7% accuracy for vision-threatening DR, 2,412 referrals; a fifth of images rejected.
94.7% accuracy for vision-threatening DR, 2,412 referrals
A third company linked to suspected money laundering
Category not assignedBank F’s network tool found phone numbers and email addresses shared by three companies. Further account checks indicated possible money laundering.
even if 10% were unrecoverable the tool contributed around USD 50 million in savings
Nuru may have lifted cassava yield from 0.5 kg to 2–5 kg a plant
Category not assignedPlantVillage's own 23 Mar 2020 post: after 18 months using Nuru, introduced via Self Help Africa and Ugunja Resource Center, Arisat reports 2.0–5.0 kg per plant against 0.5 kg; no independent source.
2.0–5.0 kg per plant against 0.5 kg
TCU's Alice flags DNIT's BR-153 notices; TCU cancels the tender
Category not assignedAlice flagged DNIT Goiás notices for BR-153 maintenance on concession stretches (R$39m per Inforchannel); TCU Acórdão 1116/2017 canceled the R$20m tender (JOTA).
ordered cancellation of the BR-153 tender budgeted at R$20 million
UCSF neuroprosthesis lets Ann Johnson speak through an avatar
Category not assigned253-electrode ECoG and deep-learning decoder let Ann Johnson, mute since a 2005 stroke, produce text, synthesized speech and an avatar
decoded text at a median 78 words per minute with 25% median word error rate
Stanford neuroprosthesis decodes Pat Bennett's speech at 62 wpm
Category not assignedFour 64-electrode microelectrode arrays and an RNN with a 125,000-word language model let Pat Bennett (ALS) produce text at 62 wpm, 23.8% word error rate
produce text at 62 wpm, 23.8% word error rate
UCSF neuroprosthesis lets Pancho produce sentences at 15.2 wpm
Category not assigned128-electrode ECoG and deep-learning decoder let a man mute since 2003 produce sentences at 15.2 wpm from a 50-word vocabulary
sentences at a median 15.2 words per minute with a median word error rate of 25.6%
ALERTCalifornia alert sends Cal Fire engines before any 911 call
Category not assignedAI camera alert at 2:33 a.m. on 6 July 2024 sent Cal Fire engines to a fire near Auburn before any 911 call; held to a 30 x 30 ft area
crews put the fire out before it spread beyond a 30 x 30 foot area
Reykjavik University ASR drafts Althingi speech transcripts
Category not assignedReykjavik University's ASR has produced Althingi's first-draft speech transcripts in production since January 2019 (Interspeech 2019).
transcribers graded 77% of transcripts as Good
Google Flood Hub triggers pay $105 to 4,600 people in Kogi
Category not assignedGoogle Flood Hub forecast triggers paid $105 to 4,600 people in Ibaji, Kogi State, mostly before floods peaked on 21 October 2024
weekly incomes rising from ₦4,500 to ₦10,000 and food shortages falling from 97% to 7%
Tesla Autopilot brakes and steers clear of a wrong-way driver
Category not assignedBret Peters's Model S on Autopilot braked and moved right as a wrong-way car came at him on Highway 49, NC; TeslaCam recorded the near miss.
Turkish Brain Project algorithm flags a glioma mid-scan at Gazi
Category not assignedReal-time MRI algorithm at Gazi University Hospital alerted doctors to a 46-year-old woman's glioma mid-scan; biopsy next day, Oct 2022.
ALERTCalifornia AI flags smoke; fire held under a quarter acre
Category not assignedAI on Cleveland National Forest cameras flagged smoke at about 3 am on 5 August 2023; Cal Fire held the verified fire under a quarter acre.
Cal Fire held the verified fire under a quarter acre
Facial recognition in Adana identifies a wanted robbery convict
Category not assignedAdana police stopped a man with no ID; facial recognition identified him as a robbery convict wanted to serve a 10-year sentence.
a fugitive convict wanted to serve 10 years, 1 month and 10 days for robbery
Taoyuan's AI sound cameras pinpoint a noisy BMW within a day
Category not assignedTaoyuan's AI sound cameras pinpointed a white BMW noise car the day after residents complained; officers caught it on 23 April 2025.
Met LFR flags David Cheneler, a sex offender, alone with a child
Category not assignedMet LFR flagged David Cheneler, 73, with a six-year-old and a lock knife at Denmark Hill, Jan 2025, breaching his SOPO; jailed 20 May 2025.
Kingston Crown Court jailed him for two years
FSD v12 may have driven its owner 13 miles to a VA emergency room
Category not assignedBy his own X account, FSD v12 drove Max Paul Franklin 13 miles to a VA emergency room at 2 a.m. on 2 April 2024; uncorroborated.
drove Max Paul Franklin 13 miles to a VA emergency room
Taichung EPB's AI smoke detection catches a serial waste burner
Category not assignedTaichung EPB AI smoke detection flagged roadside waste burning in Dadu District; a 74-year-old repeat offender was caught and fined.
a fine of up to NT$100,000
Hanoi traffic police trace a lost 78-year-old on AI cameras
Category not assignedHanoi police traced a lost 78-year-old on AI traffic cameras on 15 December 2025; a passer-by found him at 3 a.m. the next day and took him home.
Garmin Autoland lands depressurised King Air B200; pilots unhurt
Category not assignedOn 20 December 2025 Garmin Autoland flew a depressurised King Air B200 to a landing at Rocky Mountain Metropolitan Airport; two pilots unhurt.
two pilots unhurt
HCMC AI cameras identify a masked burglar; police detain him
Category not assignedHCMC street AI cameras identified a masked burglar who took two laptops and 14 million VND on 23 December 2025; police held him on 24 Dec.
Istanbul police's AVCI detains 325 and closes 69 dealer groups
Category not assignedIstanbul police's AI-based AVCI system, first used 10 February 2026 in a 14-province drug operation: 325 detained and remanded, 69 groups closed.
325 detained and remanded, 69 groups closed
Pano AI cameras spot the Bear Creek Fire; held to three acres
Category not assignedUtility-funded Pano AI cameras detected smoke from the 16 June 2024 Bear Creek Fire in Douglas County, Colorado; helicopters were dispatched within 30 minutes and the fire was held to three acres.
helicopters were dispatched within 30 minutes and the fire was held to three acres
Southern Water's sewer monitors flag a fatberg before overflow
Category not assignedSouthern Water's AI sewer monitors flagged a forming fatberg in Rowner, Gosport; a crew cleared it before any overflow into the River Alver.
Wiz finds a critical GitHub RCE with Claude Code and IDA MCP
Category not assignedWiz's AI-aided reverse engineering found a critical GitHub RCE (CVE-2026–3854); github.com was patched the same day, no exploitation found.
went from idea to working exploit in under 48 hours
CBP facial comparison at Dulles flags a French-passport impostor
Category not assignedCBP's new facial comparison system at Dulles reported a mismatch for a man presenting a French passport on 22 August 2018; a Congolese ID was found in his shoe and he was returned to Brazil.
by September 2021, more than 900 impostors nationally
ZeroEyes alert sends FSU police to a man pointing a stolen Glock
Category not assignedAt about 2:30 a.m. on 2 May 2026 ZeroEyes flagged a man pointing a gun in FSU's St. Augustine garage; officers found a stolen Glock 19 on Tyler Farmer and charged him, per court documents.
ZeroEyes alert sends Hobbs police to clay shooters, one charged
Category not assignedOn 30 November 2023 ZeroEyes alerted Hobbs police to a gun on a camera at Del Norte Pool's lot; officers detained people shooting clays near a street and charged one, per the police chief.
charges were filed against one
Cheshire facial recognition match jails county lines dealers
Category not assignedCheshire Constabulary's facial recognition matched CCTV of an unknown Northwich dealer to Keenan Hughes; he and Vincent Potter were stopped on 17 May 2024 with crack cocaine and jailed on 1 July 2024.
jailed Hughes (18) for 28 months and Potter (19) for 34 months
Tesla Autopilot stops on I-75 shoulder; drunk driver charged
Category not assignedOuest-France (3 May 2026, citing ABC 7): around 2 a.m. on 24 Apr 2026 police found a 37-year-old woman unconscious in her stopped Tesla on Interstate 75; blood alcohol over twice the limit; charged...
LNER AIVR flags a Retford track fault; overnight fix, zero delay
Category not assignedNew Civil Engineer (6 Jul 2026) quoting LNER's 9 Apr 2026 announcement: after a January 2026 Cambridgeshire defect caused over 10,000 delay minutes, One Big Circle's AIVR identified a developing fa...
an overnight repair with no delay to passengers and zero delay minutes
ChatGPT lists HSP for Phoebe Tesoriere; genetic testing confirms
Category not assignedA Cardiff teacher put her symptoms into ChatGPT after a coma; it listed hereditary spastic paraplegia, her GP called it plausible and genetic testing confirmed it in August 2025 (BBC, WalesOnline).
Foncier innovant finds 140,000 undeclared pools
Category not assignedMinister Cazenave (20 March 2024): DGFiP's AI tool detected 140,000 undeclared pools, c. €40m extra property tax for local authorities; Dordogne found 1,500 in 2023. No commune-level figure found.
detected 140,000 undeclared pools, c. €40m extra property tax
UNEP MARS alert stops Kazakhstan leak; emissions seen again
Category not assignedUNEP IMEO case KAZ_S_136: pipeline leak first detected Sept 2025, operator rerouted gas, mitigation reported 26 Nov 2025, last emissions 12 Mar 2026; subsequent emissions seen as of 21 May 2026.
estimated emissions of 1,200 tonnes
UNEP MARS alert leads to Alberta methane leak repair in a week
Category not assignedTrade press relaying UNEP's August 2026 snapshot: a 175 t/week methane plume at Alberta's Aerial field in April 2026 was traced to a failed ball valve on pigging equipment and repaired within a week.
released 175 t of methane (4,900 tCO2e) in a single week
Google's AEA alerts 16 million+ users to Marmara Ereğlisi quake
Category not assignedAEA detected the 23 April 2025 Mw 6.2 Marmara Ereğlisi quake in 5.31 s and alerted 16 million+ users (median 56 s warning for weak shaking); Google-authored evaluation; inland Take Action shortfall.
alerted 16 million+ users (median 56 s warning for weak shaking)
Skylight alerts may have led PNA to intercept XINRUN 579
Category not assignedPNA intercepted XINRUN 579 fishing illegally in Argentina's EEZ on 5 January 2025; Skylight says its speed-range and ML fishing alerts prompted the response and a first electronic fine was paid.
Tesla Model 3 steadies a Belgian baker's swerve around a wild boar
Category not assignedBaker Kenneth Van Hoesen swerved around a wild boar at 4 am on 30 May 2020 in Houthalen-Helchteren; he says his Model 3 then took over and straightened the car; Musk replied on Twitter.
Instagram self-harm alert brings Satna police to a 17-year-old
Category not assigned3 November 2022, about 5 pm: Instagram emailed MP cyber police a self-harm alert; Bhopal and Indore teams traced a 17-year-old to Satna in about an hour; she was counseled and sent home.
Rekor plate-pattern analysis leads to a Scarsdale drug stop
Category not assignedWestchester County PD's Rekor system flagged David Zayas's driving pattern across 480 cameras; a March 2022 stop in Scarsdale found crack cocaine, cash and a pistol; he was convicted (Carscoops).
Facial recognition finds a wanted woman in downtown Cuiabá
Category not assignedDiário de Cuiabá (14 July 2025): Sesp-MT's Vigia Mais MT cameras matched a woman with a June 2024 trafficking warrant in central Cuiabá on 12 July; Ciosp tracked her live and the 1st BPM arrested her.
AI cameras name 18 armed teenagers in Việt Trì within half an hour
Category not assignedTuổi Trẻ (14 June 2026): on the night of 13 June Phú Thọ police saw a group preparing weapons to fight in Việt Trì; the AI camera system identified all 18 faces in about 30 minutes; all confessed.
identified all 18 faces in about 30 minutes
Waymo pulls a car over; police find a loaded ghost gun with two minors
Category not assignedLA Times (12 September 2026): before 4 a.m. on 3 Sep Waymo pulled a robotaxi over in the Richmond District after detecting a firearm violation; SFPD found a loaded ghost gun and booked two minors.
Robot dog checks armed man on Hollywood bus; standoff ends in arrest
Category not assignedLAPD deployed its Boston Dynamics Spot quadruped during a 8 November 2023 standoff with an armed man asleep on a Hollywood Metro bus; he woke, surrendered and was arrested; the gun was a BB gun.
Kunshan police AI flags gambling ring and freezes fraud transfers
Category not assignedKunshan police say their “AI police” models led a 10 September 2024 bust of a mobile gambling ring (20-plus arrests) and froze 500,000 yuan of a 980,000-yuan telecom fraud within 10 minutes.
traced three tiers of card transfers and froze 500,000 yuan within 10 minutes
Huddersfield's first facial recognition vans: three arrests, six checks
Category not assignedWest Yorkshire Police reported three arrests and six Sexual Harm Prevention Order checks in Huddersfield on August 12–14, 2026. Its register records nine alerts and an estimated 12,875 faces seen.
three arrests and six Sexual Harm Prevention Order stops
Bradford adds multilingual, out-of-hours social-care advice
Reported benefitBradford reports that Annie handles about 50 conversations weekly, with 20% outside office hours and 74% user satisfaction.
SpeciesNet supports ecological estimates with some species missed
Reported benefitResearchers used MegaDetector and SpeciesNet to produce ecological estimates similar to expert analysis for many species, while some species were missed or misclassified.
Mean occupancy differences of 0.03-0.07 across study areas
Leeds reports more planning applications processed with Xylo
Reported benefitLeeds reports over 40% more applications processed per officer, less research time, and improved officer wellbeing during its Xylo pilot.
RFdiffusion scaffolds help solve four small-protein structures
Category not assignedA Shandong University team used RFdiffusion and ProteinMPNN scaffolds in a trimbody fiducial to solve four small-protein structures at 2.29 to 2.62 angstroms.
four small-protein structures at 2.29 to 2.62 angstroms
Face2Gene's DeepGestalt changes geneticists' testing in 17 cases
Category not assignedRudy et al. (Am J Med Genet A, online 4 July 2025) report 17 cases/19 patients in which DeepGestalt changed geneticists' testing decisions; full text inaccessible, so no per-patient detail.
17 cases involving 19 patients
MediC and FiCRO report faster clinical-trial preparation
Reported and intended benefitsMEDIAIPLUS reports 85% less trial-data alignment time, 10% lower outsourcing costs, and a 25% improvement in trial success across early deployments.
Leigong links TCM knowledge with cultivation and production
Reported and intended benefitsThe Leigong submission reports faster herbal sorting, lower Poria return costs, and improved knowledge access, alongside proposed health benefits.
PANDA reports strong pancreatic-cancer detection in CT studies
Reported and intended benefitsPANDA reports high pancreatic-lesion detection performance across large CT cohorts, including higher sensitivity and specificity than average radiologists.
Baguan reports improved regional and global weather forecasts
Reported and intended benefitsAlibaba reports lower weather-forecast error than ECMWF-IFS, including a 57.8% regional improvement, and names State Grid Shandong as a test setting.
AI maps mangrove biomass and carbon stocks in West Africa
Reported and intended benefitsdata354 describes mangrove pilots that combine satellite imagery and field plots to produce biomass maps and carbon estimates.
An LLM builds a 38,000-event natural-hazard database
Reported and intended benefitsHSE reports an operational pipeline that extracted a 38,000-event natural-hazard database from eight million Russian-language articles.
WIMPLE enables recycling collection in Busan public spaces
Reported and intended benefitsSeoreu reports that WIMPLE enabled cleaner recycling collection in previously unsorted Busan public spaces and improved collection planning.
AI Guardian supports removal of illegal wildlife listings
Reported and intended benefitsIFAW and Baidu report that AI Guardian screened 360,000 images and helped remove 7,853 illegal wildlife postings by mid-2024.
Cloudcasting improves a solar-forecast benchmark
Reported and intended benefitsOpen Climate Fix reports a 5% accuracy gain at a two-hour forecast horizon for a cloudcasting component awaiting Quartz Solar integration.
PECIP applies AI to polar monitoring and carbon-capture research
Reported and intended benefitsPECIP describes AI-assisted polar analysis and microalgae cultivation, including a claimed 30% carbon-absorption gain without a specified experimental baseline.
A telecom AI model screens content against provider policy
Reported benefitChina Telecom claims improved screening of political and ideological content; its reported classification performance does not establish a public-welfare benefit.
CloudTop expands quality inspection and customer-service support
Reported and intended benefitsChina Telecom Cloud reports quality-inspection coverage rising from 10% to 100%, 24% better service efficiency, and 30% fewer night shifts.
Telecom agents improve configuration checks and fault operations
Reported and intended benefitsChina Telecom reports configuration compliance above 95% and fault detection shortened from ten minutes to one through coordinated AI agents.
Hui Xin improves transactions in two telecom service halls
Reported and intended benefitsChina Telecom reports over 50% fewer key process errors and over 30% better marketing success in an assistant deployed in two service halls.
AI and 5G expand marine-reserve monitoring in Xiamen
Reported benefitChina Mobile reports monitoring expanding from 120 to 330 square kilometers, 60% wider wildlife distribution, and 97% fewer collision-related animal deaths.
AI-assisted MTN operations report better network performance
Reported benefitMTN attributes lower packet loss and delay to AI-assisted operations combined with network-routing changes, without isolating the model’s contribution.
Telecom AI reports large energy and operating-cost savings
Reported benefitChina Telecom reports 900 million kWh in annual savings, over 500,000 tons less CO₂, and 90% lower energy-system deployment and operating costs.
PixVerse expands a creator’s reach and supports AI filmmaking
Reported benefitPixVerse reports that creator @nerdsoul gained 300,000 followers and 210 million views in two months, alongside a separate AI-film festival example.
A deepfake model reports 95% accuracy and police applications
Reported and intended benefitsHangzhou Dianzi University reports roughly 95% deepfake-detection accuracy and applications with public-security authorities.
Safety Talk improves specialized workplace translation
Reported and intended benefitsCrowdworks reports over 90% translation accuracy and user satisfaction in multilingual field testing of construction-focused Safety Talk.
AI voice tools help people with ALS communicate
Reported benefitThe Scott-Morgan Foundation reports that its communication platform enabled Erin Taylor, who has ALS, to introduce Lenovo's CEO at Tech World.
Morphing-BOT inspects water pipelines
Reported and intended benefitsMorphing I reports that municipal pilots inspected more than 30 km of pipelines and produced validated inspection data. Avoided leaks and failures remain unmeasured.
An AI agent supports power-grid dispatch
Reported and intended benefitsChina Southern Power Grid reports that its flow-control agent outperformed dispatchers in control evaluations. Energy and emissions benefits are not quantified.
Port AI checks alarms and spots safety risks
Reported and intended benefitsZTE reports that its Taicang Port system eliminated 92% of false fire alarms and detected unsafe behavior. Prevented injuries remain unestablished.
IndustryGPT improves an industrial model-training example
Reported benefitSmartMore reports that an IndustryGPT example reached 100% accuracy in 5:27, compared with 50% in 9:31 for manual training. These are technical test results.
Reinforcement learning controls fusion plasma
Reported and intended benefitsNext Step Fusion reports that reinforcement learning controlled plasma parameters at DIII-D during 2023–24 and improved controller training. Energy benefits remain future-facing.
CreditConnect-EV expands electric-motorcycle financing
Reported and intended benefitsAIZEN reports that CreditConnect-EV financed more than 20,000 electric two-wheelers in Indonesia. Borrower income and net emissions effects are not measured.
CherryPot reports improved math learning in schools
Reported benefitZEZEDU reports academic improvement after one month of CherryPot use across Korean schools. The account provides no effect size or independent outcome evaluation.
Girls use generative AI in creative and STEM workshops
Reported benefitRHGE reports that girls in Nigerian community schools created AI-assisted art and advocacy projects. The activities expand creative practice; learning effects are not quantified.
AI maps archaeological potential at Saruq Al-Hadid
Reported and intended benefitsDubai Culture and Khalifa University report that remote sensing and AI identified known excavation areas and predicted new zones at Saruq Al-Hadid.
Robotic parking offers potential garage-capacity gains
Reported and intended benefitsShanghai AI Research Institute describes robotic parking deployments and says that single-layer layouts can increase capacity by 60–80%. The percentage is potential, not a measured outcome.
OneAstronomy improves astronomy benchmark performance
Reported and intended benefitsZhejiang Lab and NAOC report that OneAstronomy outperformed comparison models on astronomy benchmarks. Discoveries and public benefits remain unmeasured.
Dubai housing gains have unclear achieved-versus-expected status
Benefit status unclearMBRHE gives a housing-service reduction from 84 to 32 months and faster approvals, but describes the figures as both delivered features and expected benefits. Whether they were achieved is unclear.
AI sign-language broadcasting at the Winter Olympics
Reported benefitBaidu reports that XiLing provided round-the-clock sign-language coverage during the 2022 Winter Olympics, with over 100 million live and replay views.
Vivo adds offline AI accessibility tools
Reported and intended benefitsVivo describes on-device speech, vision, and sign-language features that work offline. Their direct accessibility benefits are reported; wider social effects are unmeasured.
CareVia reports less recording work and self-harm
Reported benefitSK Telecom reports recording time falling from three hours to five minutes and 76.5% fewer self-harm incidents. Its supplied reference does not identify the claimed intervention study.
Apollo Go adds accessible robotaxi controls
Reported benefitBaidu reports a limited rollout of audible arrival alerts, automatic unlocking, and voice controls for visually impaired passengers. Long-term mobility effects are unmeasured.
CargoWaves reports faster, more efficient logistics
Reported benefitDP World reports 30% higher cargo throughput and 20% lower CO2 per shipment through CargoWaves. AI’s contribution to the wider workflow improvement is unknown.
Divine Farmer provides agricultural guidance
Reported and intended benefitsChina Telecom reports agricultural Q&A accuracy above 90% and pest/disease recognition above 85%. These technical capabilities do not establish yield or income gains.
St Helens uses AI to match care needs and draft care plans
Reported and intended benefitsSt Helens built AI tools that match care assessments to a technology catalog and draft personalized care plans. Improved resident independence remains an intended benefit.
Lancashire uses AI traffic insights to support route planning
Reported and intended benefitsLancashire reports that AI traffic predictions, junction observations, and congestion analysis support transport decisions and bus-route planning.
The Amelia Scott reports better scans and expanded digital exhibits
Reported benefitThe Amelia Scott reports 50% faster image processing, improved scans, interactive exhibits, and volunteers progressing to paid museum roles.
AresGW finds eight new gravitational-wave candidates in LIGO O3
Category not assignedKoloniari et al. (Mach. Learn.: Sci. Technol., 27 February 2025): the ResNet pipeline AresGW found eight new O3 candidates with p_astro > 0.5 (cumulative 5.94); none is confirmed by another pipeline.
eight new O3 candidates with p_astro > 0.5 (cumulative 5.94)
Somerset reports time savings and improved staff focus
Reported benefitSomerset reports two to four hours saved weekly per Copilot user, with 75% reporting better focus and 53% finding work more enjoyable.
Swindon reports cheaper, faster Easy-Read documents
Reported benefitSwindon reports that Simply Readable creates Easy-Read documents in under 30 minutes at much lower cost, with input from residents with learning disabilities.
Hertfordshire trials an AI robot that repairs road cracks
Reported and intended benefitsHertfordshire reports that its ARRES pilot fixed a small number of road cracks. Long-term repair quality, safety, and cost benefits remain unmeasured.
Buckinghamshire reports time savings and better deaf access
Reported benefitBuckinghamshire reports that Copilot shortens call wrap-up and supports a deaf colleague in participating in meetings more independently.
West Midlands uses near-miss data to change road layouts
Reported and intended benefitsTransport for West Midlands reports 61,000 detected near misses and road-layout changes informed by AI. Avoided injuries are not established.
Wigan reports time savings and newly usable survey responses
Reported benefitWigan reports 107 staff days saved on minutes and new analysis of 4,500 tenant responses through QuickAction tools.
Neuromotus EMG decoding restores thumb and wrist movement
Category not assignedA 72-year-old with a paralyzed replanted arm regained active thumb and wrist movement after 18 sessions of ML-decoded motor execution.
active thumb extension (4 degrees) and active wrist flexion (6 degrees)
Johns Hopkins LSTM BCI lets a man with ALS spell for 90 days
Category not assignedA chronically implanted high-density ECoG BCI with an LSTM click detector trained on under 44 minutes of data supported switch-scan spelling for 626 minutes over 90 days, all in the laboratory.
median spelling rate of 10.2 characters per minute
AlphaFold-Multimer predicts Tmem81 as a fertilization factor
Category not assignedAn AlphaFold-Multimer screen predicted a trimeric sperm complex; the lab then showed Tmem81 is essential for male fertility in zebrafish and mice.
Princeton RL controller cuts DIII-D disruptions to two shots
Category not assignedAn RL controller trained on a tearing-likelihood model kept DIII-D plasmas below the instability threshold; before RL control all but one shot disrupted, after it only two did (Nature, 21 Feb 2024).
before RL control all but one shot disrupted, after it only two did
ChatGPT proposes ALMS1; Alström confirmed in two business days
Category not assignedChatGPT proposed ALMS1 for a 44-year-old with lifelong deafness, blindness, cardiomyopathy and CKD; Alström syndrome confirmed in two days.
Alström syndrome confirmed in two days
A physics tutor more than doubled students’ median learning gains
Reported benefitHarvard students who used PS2 Pal, powered by GPT-4-0613, learned more in less time than students in an active-learning class.
median learning gains over double those of the class session
Face2Gene suggests MRD7 after an 18-year misdiagnosis
Category not assignedFace2Gene suggested MRD7 for a 22-year-old misdiagnosed with Cornelia de Lange at age 4; exome sequencing confirmed a DYRK1A variant.
FOX matches or beats manual cochlear-implant fitting in RCT
Category not assignedIn a Brussels crossover RCT, newly implanted patients fitted with FOX had better speech intelligibility at 3 and 12 months; all manual-arm patients kept the FOX map; 47 experienced users improved.
100% of the manual group kept the FOX map and 82% performed better with it
An AI eye exam increased screening among young people with diabetes
Reported benefitIn the ACCESS trial, all 81 young people offered IDx-DR completed an eye exam, compared with 18 of 82 controls referred to an eye-care provider.
81 of 81 versus 18 of 82 completed screening
510's ML forecast triggers PRC shelter kits before Storm Nalgae
Category not assigned510 ML impact forecast hit the trigger on 27 October 2022; PRC installed shelter kits for 78 Aurora households before Storm Nalgae
installed shelter kits for 78 Aurora households before Storm Nalgae
Pregnant women gain knowledge and receive more recommended care
Reported benefitWomen invited to PROMPTS improved several measures of knowledge, preparation for childbirth, and reported care by 0.06–0.09 standard deviations.
raised knowledge and care-seeking indices by 0.06–0.09 SD
Coapt pattern recognition restores daily living for two amputees
Category not assignedTwo transulnar amputees given Coapt pattern-recognition prostheses on osseointegration implants regained activities of daily living.
AlphaFold models cover over 90% of the human NPC scaffold
Category not assignedResearchers combined AI predictions with electron-microscopy maps to model more than 90% of the human nuclear pore scaffold.
covers more than 90% of the NPC scaffold, roughly doubling previous structural coverage
BrainGate wireless iBCI lets T5 and T10 use a tablet at home
Category not assignedTwo men with tetraplegia used a wireless BrainGate iBCI at home for web browsing, apps and typing, with 24-hour home recording.
T5 typed 196 correct characters in 14.6 minutes (13.4 per minute)
More mothers keep listening to health messages
Reported benefitmMitra’s call scheduler reduced cumulative engagement drops by 32% compared with regular messaging alone over seven weeks.
32.0% fewer cumulative engagement drops than standard care
Google FINDER flags restaurants 3.1 times as likely to be unsafe
Category not assignedHealth departments inspected 132 FINDER-flagged restaurants in 2016–2017; 52.3% were unsafe versus 24.7% of 10,786 other inspections (odds ratio 3.06).
52.3% were unsafe versus 24.7% of 10,786 other inspections (odds ratio 3.06)
Berkeley CNN finds 72 unseen pulses from FRB 121102
Category not assignedA Berkeley CNN found 72 unseen pulses from FRB 121102 in Green Bank data, published in ApJ and announced by GBO (Sep 2018).
found 72 previously undetected pulses from the repeating fast radio burst FRB 121102
Google neural network finds an eighth planet, Kepler-90i
Category not assignedA Google/UT Austin neural network found Kepler-90i in archival Kepler data; NASA announced it on 14 December 2017 and the AJ published it on 30 January 2018.
a rocky planet about 30 percent larger than Earth with a 14.4-day orbit
Facebook ML maps cut Red Cross Malawi mapping area to 2,745 km²
Category not assignedFacebook's ML settlement maps cut Red Cross mapping for Malawi's June 2017 measles vaccination campaign (Measles & Rubella Initiative) from 94,080 km² to 2,745 km²
reducing the area to map from 94,080 km² to 2,745 km²
LFDCS glider detections trigger a NOAA Right Whale Slow Zone
Category not assignedWHOI/NOAA glider's automated call classifier detected right whales on 27 December 2023; NOAA declared a Slow Zone east of Portland, ME.
mariners route around the area or transit at 10 knots or less
Clinatec's ECoG decoder drives a four-limb exoskeleton
Category not assignedA 28-year-old with C4-C5 tetraplegia cortically controlled simulated walking and 8-DoF arm movements over 24 months (Lancet Neurology 2019).
64.0% success using the avatar at home and 70.9% with the exoskeleton in the laboratory
BioPatRec pattern recognition eases 48-year phantom limb pain
Category not assignedA 72-year-old with 48 years of intractable phantom limb pain reached pain-free periods after ML-driven phantom motor execution.
after ten weeks near-absent pain developing into completely pain-free periods
CGU, with ALICE support, avoids R$8.33m in overpriced purchases
Category not assignedWith ALICE support, CGU cut R$2.38m from a 2021 Ipanema hospital glove tender and R$5.95m from an INCA tender; R$8.33m avoided
prevented at least R$8.33 million of overpriced purchases
STF's VitorIA finds at least 247 similar appeals for Tema 1.261
Category not assignedVitorIA identified at least 247 similar appeals in the controversy behind Tema 1.261; STF Plenary found no general repercussion on 22 August 2023 (final 23 September 2023)
identified at least 247 similar appeals in the controversy behind Tema 1.261
nEmesis-guided inspections find 64% more C-or-worse venues
Category not assignedSNHD double-blind trial from 2 January 2015: Twitter-mined nEmesis flags yielded 9 vs 6 demerits per inspection and 11 vs 7 venues graded C or worse (64% more).
found 11 venues graded C or worse versus 7 for the control arm, a 64% improvement
Smartwatch ECG app confirms AF at sea; CIRM diverts the ship
Category not assignedOn 29 Sep 2023 a merchant ship with no ECG used a smartwatch ECG app under CIRM guidance; it confirmed AF within seconds, the ship diverted, hospital ECG confirmed AF and the seafarer was repatriated.
Alice audit at TCE-MG corrects notices worth R$91.7 million
Category not assignedTCE-MG audit with Alice of 2,297 Aug 2022 tenders: 60 irregularities (R$113,713,128.26); 53 alerts answered; notices corrected worth R$91,723,011.31 (TCE-MG, 27 June 2023).
the audit achieved corrections of notices worth R$91,723,011.31
Apple Watch breathing alerts send a 79-year-old to CPAP therapy
Category not assignedFukuoka case (Cureus, 26 May 2026): sleep-apnoea notifications prompted consultation; watch ECGs showed AF only after “elevated” breathing-disturbance nights; CPAP under way, no polysomnography.
Snake researchers review fewer images with MegaDetector
Reported benefitMegaDetector removed 94% of images from manual review in a Florida study, but the hybrid workflow detected fewer snake events overall.
94.0% fewer images for human review
UDK fraud models may have yielded 135, 36 and 72 cases
Category not assignedUDK/ATP ML fraud models (about 60 since 2019): 2023 figures of 135, 36 and 72 cases with revenue from three models; Amnesty (Nov 2024) calls their effectiveness questionable.
2023 figures of 135, 36 and 72 cases with revenue from three models
Apple Watch Series 3 drop from 99 to 65 bpm marks AF reversion
Category not assignedMelbourne cardiologists matched 199 Apple Watch heart-rate readings to a Holter: the watch showed a sudden 99-to-65 bpm drop at 19:29 that coincided with reversion from AF to sinus rhythm.
sudden 99-to-65 bpm drop at 19:29 that coincided with reversion from AF to sinus rhythm
Be My AI reads airport gate numbers and a check for blind users
Category not assignedPeer-reviewed interview studies record blind users completing everyday tasks with Be My AI, e.g., airport gate numbers, a handwritten check.
Drivers slowed down when signs warned of wildlife ahead
Category not assignedDrivers slowed by an average of 3 mph near warning signs linked to Crosstek’s AI in Virginia’s ADDWS pilot.
an average speed reduction of 3 mph as vehicles approached activated signs
Apple Crash Detection alerts firefighters to a Mount Aso crash
Category not assignedSETN (21 Jan 2026, citing Japanese media): the helicopter took off at 10:52 on 20 Jan 2026 and lost contact at 11:04; the fire department had already received a collision notification from a passen...
RFI seismic warning stops Rome-Naples trains in a 4.4 quake
Category not assignedeuronews (15 Jul 2025): about 20 accelerometric stations over 200 km feed software that analyzes P-waves and automatically slows or stops trains; in the weeks after activation a magnitude 4.4 quake...
a magnitude 4.4 earthquake, during which trains on the Rome-Naples line were stopped
KTÜ students' app pings phones about 30 seconds before a quake
Category not assignedKTÜ students' AI quake app alerted phones about 30 s before a Kulu quake (DHA 5.2, USGS 5.0) was felt at Turkey's parliament, 15 May 2025.
notifications had reached their phones 30 seconds before the quake
AlphaEvolve finds 48-multiplication 4x4 complex matrix algorithm
Category not assignedAlphaEvolve found a 48-multiplication 4x4 complex matrix algorithm (May 2025); Dumas, Pernet and Sedoglavic derived rational variants from its decomposition
4x4 complex-valued matrices with 48 scalar multiplications, improving on the 49
Google alerts Crete phones seconds before the M6.0 Kasos quake
Category not assignedAndroid users in Crete and the Dodecanese got Google's quake alert seconds before the M6.0 (Greek press 6.1) Kasos quake of 14 May 2025.
LMU's eBL platform finds manuscripts of a lost Hymn to Babylon
Category not assignedLMU's AI-supported eBL platform identified further manuscripts of a lost Hymn to Babylon; edition published in Iraq (online 17 March 2025)
without the eBL platform the edition could not have used 21 manuscripts in 31 fragments
Pindrop traces the Biden New Hampshire robocall to ElevenLabs
Category not assignedAfter a deepfake Biden robocall targeted New Hampshire's Jan 2024 primary, Pindrop's analysis against 120+ voice engines pointed to ElevenLabs; Hany Farid's Berkeley team reached the same conclusion.
the result came back well north of 99 percent for ElevenLabs
Envision Glasses read street signs and menus for Wes Ramage
Category not assignedGlobe and Mail, 26 December 2023: Wes Ramage of Calgary uses Envision Glasses (Google Glass plus Envision's AI) to read signs for transit, read menus, find chairs and objects and describe scenes.
TrailGuard AI transmits images of armed poachers in Kishanpur
Category not assignedDertien et al. (BioScience 2023): one TrailGuard AI unit captured poachers “carrying a gun and knives” on two dates and a poached chital; officials identified them. Arrests claimed only in releases.
TrailGuard AI sends tiger alerts within 30 to 42 seconds
Category not assignedBioScience 2023: 12 TrailGuard AI cameras ran in the Kanha-Pench corridor, Madhya Pradesh, mid-May to mid-July 2022 (591 trap nights); tiger notifications arrived 30 to 42 seconds after detection.
tiger notifications arrived 30 to 42 seconds after detection
Bradford AI may have identified the de Brecy Tondo as a Raphael
Category not assignedUgail's deep-learning comparison found 97% facial similarity with the Sistine Madonna; Cartwright Hall showed the Tondo July 2023-Jan 2024; Raphael scholars call it a copy; Art Recognition says not.
97% facial similarity with the Sistine Madonna
ChatGPT appeal letter gets a York student's £60 fine withdrawn
Category not assignedMillie Houlton, 22, asked ChatGPT to draft an appeal against a wrongly issued £60 parking fine; York City Council withdrew it.
£60 parking fine; York City Council withdrew it
Supreme Court of India publishes TERES AI hearing transcripts
Category not assignedFrom 21 Feb 2023 the Supreme Court of India transcribed Constitution Bench hearings live with an AI service from TERES; 9 of 25 hearings had published transcripts by May 2023, and almost 10 matters...
9 of 25 hearings had published transcripts by May 2023
Meta alert brings police and psychologists to a Paraguayan teen
Category not assignedMeta flagged a Paraguayan teen's post praising the Uvalde shooter; the FBI relayed it and police, prosecutors and psychologists intervened.
Google alerts Alanya phones seconds before a M5.1 quake
Category not assignedAndroid users near Alanya, Turkey, got Google's earthquake alert seconds before a M5.1 (AFAD 5.3) offshore quake on 5 January 2022.
Hürriyet put the lead at about 10 seconds
Beethoven X AI completes the Tenth Symphony; premiered in Bonn
Category not assignedAI-completed Beethoven Tenth premiered 9 October 2021 by Beethoven Orchester Bonn under Dirk Kaftan at the Telekom Forum, Bonn
Google Arts & Culture Lab recolours Klimt's destroyed paintings
Category not assignedML color reconstruction of Klimt's destroyed Faculty Paintings launched 7 October 2021 with the Belvedere, shown in its 2025 exhibition
myTax prompts had an estimated $37 million revenue impact
Category not assignedATO page: at tax time 2020 nearly 340,000 myTax users (7.5%) got real-time prompts from nearest-neighbor checks; adjustments had an estimated revenue impact of about $37 million.
adjustments estimated to have a revenue impact of around $37 million
Rijksmuseum neural networks restore The Night Watch's strips
Category not assignedRijksmuseum used neural networks to reconstruct The Night Watch's 1715-trimmed edges, displayed complete from 23 June 2021.
Sheng Siong's facial recognition catches a repeat wine thief
Category not assignedSheng Siong's facial recognition matched a wine thief to six prior thefts and flagged her return; arrest, restitution, eight days' jail.
matched a wine thief to six prior thefts and flagged her return
Anthropic flag on a shooting threat leads to a Florida arrest
Category not assignedAnthropic safeguards flagged a threat to shoot up a Florida sheriff's office; reviewers tipped police, who arrested the user; no gun found.
Milton Keynes reports faster planning validation with Valon
Reported benefitMilton Keynes reports faster validation and decisions during its Valon pilot, with receipt-to-validation time falling from 15.8 to 7.6 days.
Lunit INSIGHT CXR spots subpulmonic effusion; ultrasound agrees
Category not assignedAt a basic emergency service in the Algarve, the AI marked bilateral effusion (97% left, 77% right); screening ultrasound confirmed a right subpulmonic effusion; the patient refused transfer and left.
Seeing AI reads printed recipes aloud for a woman with RP
Category not assignedA 62-year-old with autosomal dominant RP used Microsoft Seeing AI to read printed recipes in her kitchen (case report, 2025).
Greater Cambridge reports faster consultation summaries
Reported and intended benefitsGreater Cambridge reports that a consultation-summary task fell from 18.5 officer hours to 16 AI minutes while officers still read every response.
Ant health assistants reach patients and support medical work
Reported and intended benefitsAnt reports that its medical AI served 13 million users and improved annotation efficiency by 50%; individual health outcomes are not specified.
Dentbird automates dental prosthesis design
Reported and intended benefitsImagoworks reports that Dentbird generates patient-specific dental designs in minutes, automating labor-intensive CAD work during real-world pilots.
RuiPath shortens pathology review at Ruijin Hospital
Reported and intended benefitsHuawei reports that RuiPath reduced pathology-review time at Ruijin Hospital, including initial review from 20–40 to 10–15 minutes per patient.
An organoid AI platform targets faster drug research
Reported and intended benefitsThe organoid platform reports early partner testing and claims that AI screening can reduce discovery timelines by 50%; treatment benefits remain intended.
GeoGPT accelerates marine-image analysis
Reported and intended benefitsGeoGPT reports marine-image recognition in 40 milliseconds versus 40 seconds manually, alongside species-classification and exploration tools.
A hydrology model reports faster operational reporting
Reported and intended benefitsZTE reports approximately 30% less hydrology report-writing time and 20% better line-drawing efficiency in its operational hydrology application.
Weather AI reduces penalties at a Huadian renewable-power pilot
Reported and intended benefitsHuadian reports penalty reductions of 27% and 15% in its Jiangsu pilot during the first half of 2024, alongside faster renewable-power forecasts.
Di Ting shortens customer-service training and handling time
Reported benefitHubei Telecom reports that Di Ting shortened new-agent training from three months to one week and reduced average handling time by 10%.
AI shortens semiconductor floor planning
Reported and intended benefitsAgileSoDA reports that partner deployments cut floor-planning time by 76%, from more than a week to roughly a day. The figures are first-party aggregate claims.
Dubai automates cement laboratory testing
Reported and intended benefitsDubai Central Laboratory reports that its robotic XRF system automates cement preparation and analysis, reducing manual work. Safety gains are not measured.
MOF IntelliServe answers queries and reads documents
Reported benefitThe UAE Ministry of Finance reports that its live assistant provides instant bilingual answers and document search, reducing routine manual effort. Cost savings remain expected.
Nego-Wiz automates KT Commerce price negotiations
Reported benefitAI O2O reports that Nego-Wiz shortened negotiations at KT Commerce during a 2024–25 pilot. The amount of time saved is unreported.
AI Payroll Guard cuts federal payroll processing time
Reported and intended benefitsThe UAE Ministry of Finance reports that its live AI-enabled payroll system reduced processing time by up to 95%. AI's share of the wider automation gain is not isolated.
AI assesses housing-loan eligibility in Dubai
Reported benefitMBRHE reports that AI credit and eligibility models automate housing-loan assessment and remove manual approval bottlenecks. AI’s separate effect is unmeasured.
QANDA offers affordable AI tutoring
Reported benefitMathpresso describes real-time question solving, adaptive feedback, and study plans through QANDA. Learning gains and reductions in educational inequality are not measured.
AI Helpy personalizes textbook feedback
Reported and intended benefitsElice describes AI textbooks with tailored feedback, language support, and teacher oversight. Learning gains and teacher time savings are unmeasured.
Dubai's Smart School Finder answers parents' questions
Reported and intended benefitsDubai's school-finder pilot provides natural-language searches using KHDA school data. Easier information retrieval is reported, while educational improvements remain unmeasured.
Tanbeeh speeds gate entry at DP World
Reported and intended benefitsDP World reports that Tanbeeh reduced gate entry from three minutes to 15 seconds. The platform also monitors fire and protective equipment; prevented injuries are unproven.
Voice commands orchestrate in-vehicle functions
Reported and intended benefitsChangan reports that its vehicle agent reduces scenario-configuration tasks from minutes to seconds. Safety and environmental benefits remain unmeasured.
South Cambridgeshire reports fewer calls and better routing
Reported benefitSouth Cambridgeshire reports fewer calls, better routing, and 86–90% approval in chatbot feedback, without evidence of improved representative resident satisfaction.
ML (VAE, custom CLADES) delimits second Micrognathozoa species
Category not assignedTranscriptomes plus machine-learning delimitation split the Crozet population as Limnognathia desmeti sp. nov. (Proc R Soc B, 19 February 2025).
West Oxfordshire and Cotswold report less planning admin
Reported benefitWest Oxfordshire and Cotswold report 85% less repetitive planning-administration time with AI tagging and summaries of consultation feedback.
Birmingham reports faster redaction and automated email indexing
Reported benefitBirmingham reports that AI redacts documents in 90 seconds instead of 10 minutes and indexes 55% of emails, saving staff time and costs.
North Yorkshire reports faster social-care data retrieval
Reported and intended benefitsNorth Yorkshire reports 90% less time and cost for some data-retrieval tasks in its Azure AI proof of concept, plus relationship-mapping tools.
Surrey Heath explores Copilot meeting and email summaries
Reported benefitSurrey Heath reports promising meeting-recap and email-summary tests with two Copilot licenses. It has not established a business case or quantified gains.
West Berkshire reports quicker job-advert drafting with GPT
Reported benefitWest Berkshire reports faster, more consistently formatted job adverts from its GPT tool, with low stated setup and running costs.
Halton introduces a family-hub digital front door
Reported benefitHalton reports that its family-hub platform offers round-the-clock information and that automation has begun reducing administrative work.
Kingston reports shorter social-care documentation with AI
Reported benefitKingston reports that Magic Notes cuts documentation time by 50–60%, with supervision write-ups falling from 40 minutes to under 10.
Barnsley reports less administrative work with Copilot
Reported benefitBarnsley reports that Copilot saves time on meeting notes, summaries, procurement, legal work, and social-care administration.
Peterborough reports quicker advice through Hey Geraldine
Reported benefitPeterborough reports 15 minutes saved per conversation with Hey Geraldine. A manager describes obtaining equipment information while a colleague was away.
Azure AI Vision app helps blind, low-vision adults in 354 tasks
Category not assignedGonzalez Penuela et al. (CHI 2024) log 354 diary entries from 16 BLV participants using an Azure AI Vision app; uses include telling sunglasses from reading glasses without calling a spouse.
2.76 out of 5 for satisfaction and 2.43 out of 4 for trust on average
FAST-PETase degrades 51 post-consumer PET products in a week
Category not assignedA 3D-CNN ensemble engineered FAST-PETase, which degraded 51 post-consumer PET products in a week; Texas A&M later reported complete depolymerisation using it with MHETase.
degraded 51 post-consumer PET products in a week
Fitbit Charge 3 signals precede 4 of 5 COVID-19 exacerbations
Category not assignedKanazawa pilot, 23 COVID-19 patients (2020): Fitbit oxygen-variation signals preceded 4 of 5 exacerbations (sensitivity 80%); one man's persistent signals led to polysomnography showing severe OSAS.
EOV signals preceded four of five analyzed cases (sensitivity 80%, specificity 90%)
Groningen AI finds two scribes wrote the Great Isaiah Scroll
Category not assignedAI handwriting analysis showed two scribes wrote the Great Isaiah Scroll, switching at columns 27–29; published in PLOS ONE 21 April 2021
two scribes wrote the Great Isaiah Scroll, switching at columns 27–29
GPT-5.5 Pro solves Erdős problem #38; solution Lean-formalized
Category not assignedGPT-5.5 Pro solved Erdős problem #38; human-checked 26 April 2026, Lean-formalized by Aristotle 1 May, recorded as solved on erdosproblems.com
BirdCast forecasts underpin Houston's Lights Out Nights
Category not assignedHouston Audubon issues Lights Out alerts from BirdCast forecasts; City of Houston proclaimed Lights Out Nights on 9 April 2021.
Oura ring trends feed into a Holter showing 2,078 PVCs in 24 h
Category not assignedDubai clinic case: Oura ring heart-rate and HRV summaries, read by clinicians, fed into a stress test and Holter showing 2,078 PVCs in 24 h; the ring detects no arrhythmia and issued no alert.
a 24-hour Holter recorded 2,078 ventricular events (1.75% of beats, no couplets or runs)
Universal Credit advances model saves an estimated £4.4 million
Category not assignedThe National Audit Office reports that DWP's model, in use since May 2022, flags high-risk advance requests for caseworker review, with estimated savings of £4.4 million and about three times the h...
saving an estimated £4.4 million from 2022–23 to 2024–25 by DWP's estimate
Study suggests seven-day flood warnings in parts of Mali
Category not assignedForecasts could allow earlier action in parts of Mali, but Google Flood Hub produced frequent false alarms. No benefit from an implemented early action was measured.
skillful beyond 4-day lead time, enough for 7-day Mali Red Cross early action triggers
ECMWF's AIFS Single goes operational with 5–15% lower errors
Category not assignedECMWF's AIFS Single became operational 25 February 2025; ECMWF reports 5–15% lower errors than IFS; no specific event documented.
5–15% lower errors than IFS
SoundTalks health score falls with high swine influenza A loads
Category not assignedLMU Munich blinded study in a German nursery: swine influenza A loads in oral fluids and bioaerosols were associated with falls in the AI's respiratory health score; no intervention outcome.
GraphCast matches best operational cyclone track models in 2023
Category not assignedCIRA/Colorado State evaluation: GraphCast and three AI models match top operational cyclone track models in 2023; adding all four cut consensus error 11% at 120 h.
adding all four improved the NHC track consensus by 11% at 120 h
ADS crash odds are lower in most scenarios, higher at dusk
Category not assignedNature Communications (18 Jun 2024): using California DMV, NHTSA Standing General Order and SWITRS data, Abdel-Aty and Ding find ADS crash odds lower in most matched scenarios but 5.25x higher at d...
Compared with human-driven vehicles, the odds of an ADS crash in rain were 0.335 times
GNoME's 2.2 million predicted crystals may include useful compounds
Category not assignedGoogle DeepMind's GNoME proposed 2.2 million crystal structures (380,000 stable); independent reviews (Cheetham & Seshadri 2024, Kurlin 2026) find duplicates and scant evidence of useful compounds.
2.2 million crystal structures (380,000 stable)
Nearly a third of DGFiP tax controls rest on mass-data analysis
Category not assignedThe Cour des comptes (November 2023) found that nearly a third of DGFiP tax controls rest on mass-data risk analyzes (target 50% by 2027); a July 2026 parliamentary exchange put data mining at 44% ...
nearly a third of DGFiP tax controls rest on mass-data risk analyzes (target 50% by 2027)
IRS applies AI to select large partnerships for examination
Category not assignedIRS (8 September 2023): AI to help detect cheating and select cases; exams planned for 75 partnerships averaging over $10bn in assets. GAO (Jun 2024): two AI models prioritize partnership audits.
YES-TECH may speed PMFBY crop-insurance claims
Category not assignedLaunched 21 July 2023; nine states by Jan 2025; Madhya Pradesh 100% weightage from kharif 2023. PIB claims faster claims; CSE and DTE report lower payouts and rollback calls in three districts.
kharif payouts in the three districts fell from Rs 217.91 crore to Rs 55.2 crore
DGFiP's Foncier innovant leads to 20,000+ pools taxed for 2022
Category not assignedDGFiP deep-learning pool detection: 20,000+ pools taxed for 2022 (~€10m for communes); 150,000 taxable pools found across metropolitan France excluding Paris and overseas departments
more than 20,000 pools being taxed for 2022, close to €10 million of extra revenue
Ithaca raises historians' accuracy from 25 to 72 percent
Category not assignedIthaca restores and dates Greek inscriptions with measured accuracy gains; no specific accepted restoration verified.
raised historians' accuracy from 25 to 72 percent when they worked with it
Exomiser helps diagnose 25% of 2,183 probands in the NHS pilot
Category not assignedPhenotype-driven ML prioritization helped diagnose 25% of 2,183 probands in the NHS pilot; a quarter of diagnoses were immediately actionable.
Genetic diagnoses were found in 25% of probands, rising to 35% for monogenic disorders
STF's Victor OCR processes 10 million+ pages; classifier pending
Category not assignedSTF–UnB Victor: OCR layer live since Dec 2020 (10m+ pages by May 2021); splitter and repercussão geral classifier not yet in production (STF, 19 August 2021); time savings were projections.
more than 10 million pages processed by May 2021
IdentiFlight cuts eagle deaths by an estimated 50–82%
Category not assignedPeregrine Fund, WEST and USGS researchers estimated an 82% (75%–89%) relative cut in eagle deaths at a Wyoming wind farm after IdentiFlight curtailment; a 2023 reanalysis put it at 50% (−159%, 89%).
82% (75%–89%) relative cut in eagle deaths at a Wyoming wind farm
GLAD alerts cut deforestation probability in Africa by 18%
Category not assignedIndependent evaluation: subscriptions to GLAD Landsat alerts (bagged decision trees) cut deforestation probability in Africa by 18%.
decreased the probability of deforestation in Africa by 18 percent
PlantVillage Nuru out-diagnoses extension agents, 65% vs 40–58%
Category not assignedIITA-led evaluation in Kenya and Tanzania: the offline deep-learning app out-diagnosed extension officers (40–58%) and farmers (18–31%); six-leaf assessment raised accuracy to 74–88%.
65% accuracy in 2020 versus 40–58% for extension agents and 18–31% for farmers
Dytective online dyslexia screener reaches 80.6% recall
Category not assignedRello et al.'s random-forest screener, tested on 3,644 children from 113 organizations in five countries, 79.8% accuracy; the tool has been used over 200,000 times. A 2026 paper reports confounding.
79.8% accuracy with 80.6% recall and 79.3% precision
Corte Constitucional's PretorIA remains in user testing in 2021
Category not assignedPretorIA, presented by the Corte Constitucional on 27 July 2020 to support tutela selection (620,242 tutelas in 2019), was in user-testing in Apr 2021 (Dejusticia); no quantified outcome found.
MIT neural network finds halicin, active in vitro and in mice
Category not assignedA deep-learning screen identified halicin, active in vitro and in mice and confirmed by other labs; no human outcome.
Dexcom G6's Urgent Low Soon alert may cut time below 54 mg/dL
Category not assignedDexcom-authored analyzes (1,424 and 15,000 users) link the 20-minute predictive low alert to less hypoglycemia; an independent 24-person crossover RCT found a reduction only below 2.8 mmol/L.
time below 54 mg/dL fell from 1.0% to 0.6%
FCW and AEB cut rear-end striking crash rates 27–50%
Category not assignedIIHS: FCW alone, low-speed AEB and FCW+AEB cut rear-end striking crash rates 27%, 43%, 50% in 22 US states, 2010–2014 (cohort)
reduced rear-end striking crash involvement rates by 27%, 43% and 50%
Insilico's rentosertib improves FVC in a phase 2a IPF trial
Category not assignedPhase 2a GENESIS-IPF trial in China: 60 mg daily rentosertib +98.4 mL FVC vs -20.3 mL placebo at 12 weeks; cohort-level, sponsor-run
mean forced vital capacity change of +98.4 mL versus -20.3 mL on placebo over 12 weeks
Google flood alerts in Bihar cut reported illness by 16.6%
Category not assignedBihar RCT, 319 panchayats: households relayed Google flood alerts reported 16.6% fewer illness symptoms and about 30% lower medical costs.
16.56% fewer illness symptoms and about 30% lower medical expenditure
OrCam readers raise daily-living test scores from 2.5 to 9.5
Category not assignedIndependent cohorts (12, 100, 6 and 10 participants) show OrCam devices raising reading and daily-task scores; no individual outcome.
improved on a 10-item daily-living test from 2.5 to 9.5
SUVAS translates 36,271 Supreme Court judgments into Hindi
Category not assignedSUVAS translated 36,271 Supreme Court judgments into Hindi and 17,142 into 16 languages by 5 August 2024; High Courts report own totals
translated 36,271 Supreme Court judgments into Hindi and 17,142 into 16 languages
HMRC's Connect yields about £4.6 billion of extra tax in 2024–25
Category not assignedHMRC's Connect data-matching system yielded ~£4.6bn extra tax in 2024–25 per FOI; C&AG confirms its use to stop risky repayments
£3.4 billion of additional annual yield, about £4.6 billion in 2024–25
Observe's AI registers salmon activity rising with gill disease
Category not assignedUniversity of Stirling study, 2023 season: fish activity rose from 25.6% to 43.6% (Farm A) and 24.9% to 32.6% (Farm B) after gill health deteriorated; proliferative gill disease drove it.
fish activity rose from 25.6% to 43.6% (Farm A) and 24.9% to 32.6% (Farm B)
Whale Safe ratings support ship slow-downs; cooperation 59–85%
Category not assignedLaunched 17 Sep 2020 by UCSB's Benioff lab; grades ships' cooperation with NOAA's 10-knot zones. KCBX: vessels slowed 59% of the time in 2023; BWBS 2024 report: 85% cooperation. No vessel outcome.
an 85 percent cooperation rate overall by fleets
GraphCast may have forecast Lee's landfall nine days ahead
Category not assignedDeepMind says a live GraphCast run on ECMWF's site forecast Lee's Nova Scotia landfall about nine days ahead; NHC records landfall 16 September 2023; no forecaster use of the run is documented.
forecast Lee's Nova Scotia landfall about nine days ahead
UNEP records eight methane mitigations in Turkmenistan
Category not assignedUNEP lists eight mitigation cases, with site identifiers and dates from November 2024 to March 2026. The contribution of AI is described at system level, rather than isolated for each repair.
eight listed mitigation cases; 192 alerts and about 20% answered in the August 2026 report
Transport for NSW trials PAWS wildlife cameras; no results yet
Category not assignedTransport for NSW's Protective Animal Warning System: two AI-enabled roadside cameras on a 13 km Monaro Highway stretch trigger driver warnings; trial announced 18 July 2026, no results yet.
WhaleSpotter thermal AI starts whale alerts in San Francisco Bay
Category not assignedWhaleSpotter’s thermal AI and human review feed whale alerts to San Francisco Bay mariners; independent reporters documented deployment and Coast Guard use, without a measured decline in strikes.
Greece presents AI to transcribe and check GSM-R rail dialogues
Category not assignedIn a SKAI interview, Greece’s transport minister described AI transcription of GSM-R conversations and checks against railway communication protocols; no measured safety result was given.
Amazon Mining Watch maps 496,000 ha of mining in protected areas
Category not assignedAI mining detector (Earth Genome, Amazon Conservation, Pulitzer Center) maps mining scars basin-wide; 496,000 ha in protected areas 2018–2025 reported Apr 2026; no attributable enforcement event.
496,000 hectares affected between 2018 and 2025
Orca AI may have cut high-severity close encounters by 52%
Category not assignedRiviera and Hellenic Shipping News (16 Apr 2026): the P&I club and Orca AI analyzed 10.8 million nautical miles; high-severity close encounters fell from 2.7 to 1.3 per vessel-month between months ...
high-severity close encounters fell from 2.7 to 1.3 per vessel-month
Ukrainian robots and drones may have taken a Russian position alone
Category not assignedZelenskyy (13 April 2026): an enemy position was taken “exclusively by unmanned platforms” with no infantry and no losses, per Politico and Business Insider; no location, date or verification.
more than 22,000 missions on Ukraine's front line
Lime fits Lime Vision sidewalk alerts to 3,500 Seattle scooters
Category not assignedSeattle Times (20 March 2026): Lime Vision uses cameras and AI to detect sidewalk riding and emit repeating audible alerts; 3,500 scooters by 1 June 2026, whole fleet by year-end; no outcome data yet.
Iron Beam may have made its first combat interception
Category not assignedYnet and JPost (2 March 2026) call the video unconfirmed; Times of Israel (27 Mar, 30 June 2026) quotes officials that Iron Beam “has not yet been declared operational”. No cited source calls it AI.
one drone intercepted near Shlomi
Metalesa AI pilot lights wild-boar warning signs on the CV-600
Category not assignedEuropa Press (12 Feb 2026): a solar-powered system combining cameras, computer vision and detection algorithms lights warning signs only when animals are present; the provincial council will evalua...
AI crossing cameras to alert train drivers at Nonsan and Boseong
Category not assignedKyunghyang Shinmun (11 Feb 2026): MOLIT's crossing-accident plan from March 2026 includes AI intelligent CCTV piloted in Q1 at two crossings where 2025 accidents occurred; 36 crossing accidents in ...
ProRail uses Genetec analytics on 7,000 cameras to flag trespass
Category not assignedSpoorPro: ProRail swaps 288 lineside cameras for smart cameras in 2026, tests drones and runs Genetec video analytics on 7,000 cameras to flag trespassers to dispatchers; no results are given.
GRA's Publican AI may have recovered GH¢15 million
Category not assignedGRA's Publican AI customs analytics, supplied by an unnamed Cyprus-based firm, rolled out from 1 February 2026 at Tema Port; the OECD AI Incidents Monitor summary says GH¢15 million was recovered, ...
the system has already recovered GH¢15 million from fraudulent importers
RBIH's MuleHunter.AI adopted by 23 banks; RBI withholds figures
Category not assignedMediaNama's RTI (December 2025) found 23 banks had implemented RBIH's MuleHunter.AI, which analyzes transaction patterns to spot mule accounts; the RBI refused to release detection figures, citing ...
An OCT-reading model beats ChatGPT-5 at planning coronary stents
Category not assignedarXiv preprint (11 December 2025): CA-GPT on an AI-OCT system agreed with expert PCI records more than ChatGPT-5 and junior physicians across 96 patients; stent diameter 90.3% vs 72.2% for juniors.
stent diameter (90.3% vs. 72.2%, P<0.05) and length selection (80.6% vs. 52.8%, P<0.01)
Quartz Solar goes live at NESO and may avoid £30m a year in costs
Category not assignedOpen Climate Fix's machine-learning solar forecast is live in NESO's control room; OCF claims at least £30m a year in avoided imbalance costs, a figure NESO's quoted statement does not give.
at least £30m a year in avoided imbalance costs
AIxCC systems find 54 of 63 synthetic vulnerabilities, patch 43
Category not assignedAt DEF CON 33 (8 August 2025) DARPA reported seven teams' systems found 54 synthetic vulnerabilities, patched 43, and found 18 real bugs for disclosure; Team Atlanta won $4 million.
found 54 synthetic vulnerabilities, patched 43, and found 18 real bugs for disclosure
India's DoT disconnects over 82 lakh connections flagged by ASTR
Category not assignedDoT's indigenous ASTR tool matches subscriber photos to find connections taken by one person under different names; a Rajya Sabha reply (August 2025) counts more than 82 lakh disconnected after fai...
disconnected more than 82 lakh connections through ASTR after they failed re-verification
LAARMA alerts cut fatal crashes 31% in Kuranda cassowary trial
Category not assignedTMR three-month trial at Kuranda: 97% cassowary detection, drivers slowed 6.6 km/h, fatal crashes down 31% (reported 30 July 2025).
fatal crashes fell 31 percent during the trial
FDA uses AI to target food inspections
Category not assignedCommissioner Marty Makary told a Senate committee in May 2025 that FDA uses AI to target food inspections; a multi-year imported-seafood AI pilot is mentioned; no outcome figures in the source.
Sky Sentinel AI turret may have downed four Shahed drones
Category not assignedUkraine's Sky Sentinel, an AI-controlled heavy-machine-gun turret, is reported to have downed four Russian Shahed drones with one prototype; dates and places are withheld (May-June 2025).
ChatGPT's first-aid advice may have saved a teen in Assiut
Category not assignedA Facebook post of 24 May 2025 credits ChatGPT with first aid. Assiut Faculty of Medicine dean Alaa Attia says that the youth had no phosphine symptoms and said he threw the pill into water.
Treasury's machine learning recovers $1 billion in FY2024
Category not assignedTreasury reported over $4 billion in fraud and improper payments prevented or recovered in FY2024, including $1 billion recovered through machine-learning check-fraud detection.
$1 billion recovered through machine-learning check-fraud detection
KWS's FLIR cameras send intrusion alerts at three rhino sites
Category not assignedKWS installed AI-enabled FLIR thermal cameras at three black rhino sites by Sept 2024 (US State Department and WWF funding); intrusion alerts only; no AI-linked arrest or prevented event documented.
AlphaProof and AlphaGeometry 2 reach IMO silver-medal standard
Category not assignedDeepMind AI scored 28/42 on IMO 2024 problems, graded by Gowers and Myers; Gemini Deep Think's 35/42 in 2025 was IMO-certified gold
awarded 28 of 42 points, a silver-medal score
HALO's AI may spot mines in Ukraine with about 70% accuracy
Category not assignedHALO announced a US$4m AWS package (25 June 2024) to trial AI on drone imagery of Ukrainian minefields; Reuters (30 April 2026) reports about 70% accuracy; no cleared land or found mine attributed.
achieving around 70% accuracy already
Kota police act on a Meta self-harm flag within a week of tie-up
Category not assignedKota City SP Amrita Duhan said Meta's red-flag tags for suicidal posts go to a round-the-clock police team; within a week, police prevented a possible suicide by a student from Jhunjhunu (PTI).
Te Hiku Media's te reo Māori recognition may be 92% accurate
Category not assignedIwi broadcaster's ASR, trained on 310 hours of community recordings (86% accuracy, 2022), reported at 92% in 2024; archive tests announced in 2024 with results to come; no speaker outcome documented.
transcribes te reo Māori with 92% accuracy
SES Water's AI leak alarms may cut leakage run times by up to 40%
Category not assignedSES Water installed 1,000 underground flow and pressure sensors with AI prediction that raises leak alarms within minutes; it told the BBC leakage run rate times fell by up to 40% (11 March 2024).
reduced leakage run rate times by up to 40%
ScamShield flags about 5 million SMSes by October 2023
Category not assignedMHA told Parliament in November 2023 that ScamShield, built by Open Government Products with the NCPC and the Police, had about 750,000 downloads, about 80,000 numbers blacklisted and about 5 milli...
about 5 million SMSes had been flagged to users as potential scams
Tel Aviv and Ariel models translate Akkadian at BLEU4 36.5–37.5
Category not assignedTel Aviv and Ariel University researchers trained CNN-based models translating Akkadian from cuneiform glyphs or transliteration to English, best on short sentences; a capability demonstration only.
BLEU4 scores of 36.52 and 37.47
FCW with AEB cuts front-to-rear crashes 49%
Category not assignedVOA Learning English (16 Nov 2022): the automaker-NHTSA Partnership for Analytics Research in Traffic Safety analyzed 12 million police-reported crashes in 13 states; IIHS analyzed pickups in 25 st...
front-to-rear crashes reduced by 49%
REACH VET tied to fewer suicide attempts, not deaths
Category not assignedVA-authored triple-difference cohort study (173,313 veterans): REACH VET inclusion linked to more completed visits and safety plans and fewer admissions, ED visits and attempts; no mortality effect.
documented suicide attempts (-0.05; -0.06 to -0.03)
Huawei digital twin's window alert comes from a sensor, not AI
Category not assignedA Huawei-hosted launch on 25 Feb 2021 presented a window-opening alert sent to a guard's wristband. Knews and Jiefang Daily say that a window sensor detected it; the story is retold with three dates.
995 door/window alerts from January to May 2021
RFCx Guardian alerts may have stopped logging in Solok
Category not assignedRFCx Guardian sensors detect chainsaws with AI and alert patrols; a 2021 Reuters report from Solok, West Sumatra quotes a patroller that logging stopped. No dated intervention is documented.
Kuzi's SMS alerts may warn of locusts up to three months ahead
Category not assignedSelina Wamucii's Kuzi combines satellite, soil-sensor and weather data with machine learning to map locust risk for Kenya, Uganda, Ethiopia and Somalia; maker statements only, no outcome evidence.
Optage and Sanyo Electric Railway trial AI crossing cameras
Category not assignedMynavi News (12 Aug 2020, from a press release): edge-device AI detects people including prams and wheelchairs in crossings and signals drivers via special signal lamps and an IP-VPN link to the op...
Fujitsu Zinrai trial speeds vessel warnings by about two minutes
Category not assignedThe Maritime Executive (15 Apr 2020): in a Dec 2019 to Mar 2020 trial, six experienced controllers re-ran past near-miss cases with and without AI risk predictions from AIS and radar data; warnings...
the number of warnings issued to risk-prone vessels nearly doubled
VeriPol flags 25 and 39 false robberies in Murcia and Málaga
Category not assignedThe Spanish National Police's text-based false-report classifier flagged 25 false robberies in Murcia and 39 in Málaga in a week in June 2017 (versus 3 and 12 in the same month over the prior decad...
25 and 39 false robberies respectively, against 3 and 12 in the same month
FAO and Penn State launch Nuru to confirm fall armyworm damage
Category not assignedFAO announced on 25 June 2018 a talking Android app built with PlantVillage to confirm fall armyworm damage offline and feed FAO's FAMEWS; no outcome data in the sources.
Raleigh police use Traffic Jam to link a trafficker's movements
Category not assignedCBN (18 Oct 2017) shows Raleigh PD detectives using Traffic Jam, a CMU-born tool with facial recognition over escort ads, to corroborate a trafficker's movements; outcomes stated only as aggregates.
Kyukyu VoiceTra translates for ambulance crews across Japan
Category not assignedFrom April 2017 NICT and the FDMA provided Kyukyu VoiceTra (VoiceTra with medical terms and 46 fixed phrases, 15 languages) to fire headquarters across Japan; Sapporo used it from February 2017.
NCAR's DICast cuts Xcel's wind forecast error to 10.8%
Category not assignedNCAR's adaptive forecast-blending system (DICast), used by Xcel Energy since 2009, cut mean wind forecast error from about 18% to 10.8% by 2014; Xcel estimated US$49m saved (IEEE Spectrum 2015).
cut mean wind forecast error from about 18% to 10.8% by 2014; Xcel estimated US$49m saved
Chicago's model finds critical violations 7.5 days earlier
Category not assignedDouble-blind retrodiction of 1,637 Chicago inspections (Sept-Oct 2014): the model would have found 69% of critical-violation sites in the first half of the work versus 55%, 7.5 days earlier.
69% of critical-violation sites in the first half of the work versus 55%, 7.5 days earlier
TCU's ALICE tender-analysis robot captures R$426m in 2021
Category not assignedTCU says its ALICE tender-analysis robot captured R$426m in 2021 (R$194m in 2020); specific cases reported only via press
captured R$426m in 2021 (R$194m in 2020)
Some subprime borrowers get lower-priced credit at LendingClub
Category not assignedJagtiani and Lemieux found that LendingClub rating grades' correlation with FICO fell from about 80% (2007 loans) to about 35% (2014–15) and that some traditionally subprime borrowers received bett...
about 8 percent of A-rated and 28 percent of B-rated borrowers had FICO scores below 680
Agenzia delle Entrate activates VeRa to list high-risk taxpayers
Category not assignedVeRa (Verifica dei Rapporti Finanziari) cross-matches the financial-relationships archive with tax data under pseudonymisation to build selective lists of high-risk taxpayers; no realized outcome f...
Stanford's Tutor CoPilot lifts math mastery 4 percentage points
Category not assignedRCT with 900 tutors and 1,800 grade 3–8 students in a southern US district: students whose tutors had Tutor CoPilot were 4 points more likely to master topics; 9 points for lower-rated tutors.
students of treated tutors were 4 percentage points more likely to master topics
Odakyu to brake trains when cameras see someone inside a crossing
Category not assignedITmedia (22 Jun 2026): from 24 Jun 2026 AI analyzes existing crossing cameras, tracks pedestrians, wheelchair users, cyclists and motorcyclists and sends a brake signal to approaching trains; devel...
Seibu puts OKI AI crossing anomaly detection into full operation
Category not assignedResponse.jp (11 Nov 2022): OKI and Marubeni Network Solutions' AI monitoring and a stereo-camera 3D system detect people or vehicles left in crossings, light special signal lamps for drivers and se...
Odakyu starts Panasonic fall-detection AI on station cameras
Category not assignedFrom 1 April 2020 Odakyu's system analyzes existing platform camera video to flag people who fall onto the tracks or linger at the platform edge and alerts station staff; no outcomes are reported.
OPTiM trials AI camera monitoring at two JR Kyushu stations
Category not assignedOPTiM's release says its cloud AI monitors station camera video in real time, recognizing people, trains and the platform yellow line, and alerts administrators on anomalies; no results published.
Face matches reported in a search for missing children
Category not assignedThe ministry reported 2,930 facial matches among children in care. The Delhi High Court later recorded that the software had not helped solve any missing-child case.
matched 2,930 identities on 6–10 April 2018
Microsoft Translator adds Inuktitut and Inuinnaqtun with Nunavut
Category not assignedNeural MT for Inuktut dialects built with the Government of Nunavut and volunteer speakers; Nunatsiaq News and VICE report launches and translation errors; no measured user outcome documented.
X-Guard decoy may have diverted Pakistani missiles from IAF jets
Category not assignedEvery account traces to ex-USAF pilot Ryan Bodenheimer via idrw.org (6 July 2025). No IAF, MoD, Dassault or Rafael confirmation; CDS Gen Anil Chauhan admitted on 31 May 2025 that India lost jets.
Qualcomm's humanoid may have collapsed safely at Computex
Category not assignedVendor statement via Creative Bloq (27 July 2026): “The safety systems worked as intended.” NDTV, Asianet and TechSpot describe a face-first or backward collapse; nobody is reported at risk.
Xuanwu Hospital implants China's first responsive neurostimulator
Category not assigned9 Mar 2022: Xuanwu Hospital, Beijing, implanted Epilcure, a closed-loop neurostimulator whose on-skull AI chip monitors EEG and stimulates on predicted seizures, in a man with drug-resistant epilepsy.
Researchers found 47 known gravitational-wave signals
Category not assignedResearchers used MADGRAV-2D to recover 47 cataloged gravitational-wave signals from LIGO data.
47 detections with a false alarm rate below one per year and p_astro above 0.9
North East Derbyshire explores Copilot for routine work
Intended benefitNorth East Derbyshire uses Copilot for drafting, summaries, and analysis. Reduced repetitive work remains intended; the source does not establish an AI-produced cultural benefit.
Manchester residents shape a proposed AI advice service
Intended benefitManchester residents proposed safeguards for a public-facing Caddy service. Wider access to understandable advice remains intended; the panel’s design recommendations are reported.
Caseworkers report easier reporting and case handling
Reported benefitStaff in Representative Moulton’s office report easier workflows with Civic Revere, but the interview does not separate AI from conventional software improvements.
DeterTech AI cameras may have cut trespass at Dingwall
Category not assignedNetwork Rail reported fewer trespass incidents after installing DeterTech AI cameras at Dingwall, alongside signage and campaigns. It supplied no post-installation count and noted a renewed rise.
Rapiscan CT algorithms detect marine wildlife at 86–96% in testing
Category not assignedA vendor-affiliated CT study reported 86–96% detection of three marine-wildlife classes and 13% overall false alarms; testing used original scans and synthetic composites, with no customs outcome.
detection rates of 95% (shark fin), 96% (seahorse) and 86% (sea cucumber)
ForeSight-HF scores two nursing-home heart-failure cases
Category not assignedMount Sinai authors describe ForeSight-HF (GPT-5 mini on ProviderLoop) recommending diuresis for a warm/wet and de-escalation for a cold/dry nursing-home patient; both were managed independently.
AI generates property images for design discussions
Intended benefitDIY Render’s EDIT generates and revises property images for design discussions; faster sales and measured time savings remain unverified.
Peterborough develops OSCAR to connect council AI services
Reported and intended benefitsPeterborough and Datnexa are developing OSCAR to connect council services. The council reports that OSCAR Workforce is live, but has not supplied measured results.
AI triage of abuse imagery under test to find victims faster
Category not assignedOn 13 April 2026 the Met said it is exploring AI to grade and triage child sexual abuse imagery, testing with multiple companies, to flag unknown victims faster; no deployment or outcome yet.
DeepMind's GDMI beats NHC Atlantic track forecasts at 12–72 h
Category not assignedNHC's 2025 verification: Google DeepMind ensemble mean beat official Atlantic track forecasts at 12–72 h; first AI models in NHC operations.
beat official Atlantic track forecasts at 12–72 h
Corsano CardioWatch recognizes a VF arrest in post-processing
Category not assignedRadboudumc DETECT-3 case: a 64-year-old's VF arrest while cycling was ended by his ICD; the CardioWatch PPG algorithm later detected pulselessness at 09:50:17 and the fall; no live alert was sent.
detected cardiac arrest at 09:50:17 after loss of pulse
HEART-SAFE algorithm flags a terminal arrest from Watch5 Pro PPG
Category not assignedAmsterdam UMC case: a Samsung Watch5 Pro worn after treatment withdrawal captured PPG; the HEART-SAFE algorithm, run afterwards, marked circulatory arrest in step with the ECG; the patient died.
Synthetic personas help teams explore market questions
Intended benefitCambium offers conversations with synthetic personas to help teams explore prices, messages, and product ideas; research accuracy and business benefits remain unverified.
Nephew uses ChatGPT during FOLFIRINOX; liver metastases regress
Category not assignedA dentist nephew in Baku used ChatGPT, later GPT-5, to summarize literature and weigh off-label add-ons for a 65-year-old on FOLFIRINOX; 11 liver metastases regressed; the author cannot attribute it.
11 liver metastases regressed
An AI companion is being developed for style reflection
Intended benefitYourcore’s founder describes AI-assisted app development and an intended style companion; benefits to users’ wellbeing or consumption remain unmeasured.
Voters receive an AI assistant for researching their ballots
Intended benefitVoteMate provides conversational ballot research, while improvements in voter knowledge and participation remain unmeasured in the reviewed sources.
An AI interface connects users to library collections
Reported benefitNortheastern’s library connected Claude to scholarly collections and reported better discovery in early use; learning gains have not been measured.
Generate Biomedicines' AI-designed GB-0669 completes phase 1
Category not assignedPhase 1 of the AI-designed antibody GB-0669 in 51 healthy adults showed tolerability, a 54-day half-life and dose-dependent virus neutralization.
dose-dependent serum live-virus neutralization occurred at 600 and 1,200 mg
Virginia tests AI to help agencies review regulations
Intended benefitVulcan Justinian analyzes Virginia regulations, but the reviewed reporting does not establish how much AI contributes to claimed economic savings.
GiveDirectly's MobileAid may have halved enrollment cost in Malawi
Category not assignedGiveDirectly posts (2024–2025) describe phone-metadata AI/ML targeting in Malawi: 35% of recipients enrolled via phone data, cost per enrolee halved; no date, count or independent source.
35% of recipients enrolled via phone data, cost per enrolee halved
wi.breathe cameras reportedly flag a Troizina wildfire
Reported benefitWINGS and Troizinia-Methana’s mayor credit wi.breathe with an August 2025 wildfire alert. Contemporary reporting describes spread and a substantial firefighting response.
AI animated archival photographs for a historical film
Reported benefitGiovanni Abitante used generative video to animate archival photographs for I Am an American; educational effects have not been measured.
UNDP hands Mines Eye to SESU after a pilot scanning over 360 ha
Category not assignedMines Eye (magnetometry, visual/infrared imaging, AI detection of explosive remnants) scanned over 360 ha in a five-month SESU pilot; handed over 21 June 2025. No detection or land-release counts.
scan over 360 hectares of confirmed and suspected hazardous areas
Northumberland develops FloodAI for six flash-flood catchments
Intended benefitNorthumberland describes an alpha-stage flood-forecasting service for six catchments. No issued warning or avoided flood impact is reported.
A virtual yoga instructor offers automated posture feedback
Intended benefitC-DAC presents an active yoga-instruction demo that aims to make posture guidance affordable; practitioner health effects are not quantified.
CAFEIN trains research models without centralizing sensitive data
Reported and intended benefitsCERN describes operational federated model training and analysis across institutions, with intended benefits for research and privacy-preserving collaboration.
TinyML proposes smartphone-assisted malaria microscopy
Intended benefitA Ghanaian research project proposes smartphone and TinyML support for malaria microscopy, with rural diagnostic access as its intended benefit.
Animal biometrics support registration and vaccination tracking
Intended benefitiSciLab describes horse-passport and vaccination-tracking pilots, aiming to improve animal identification, welfare, and disease surveillance.
Remote-care AI combines monitoring and triage support
Intended benefitTata Elxsi describes an early healthcare MVP for underserved communities, including controlled monitoring and shadow-mode triage.
Model-assisted ear-disease mapping aims to guide health services
Intended benefitNT Health proposes model-based disease-pattern analysis to guide ear-health services.
AI content and companions support senior-care activities
Intended benefitESTsoft reports deployment in 36 senior centers and a 500-person demonstration, while cognitive, physical, and emotional effects remain unspecified.
HIBA offers multilingual health information in Lebanon and Jordan
Reported and intended benefitsHIBA was tested with community groups in Lebanon and Jordan to improve access to understandable health information in text and audio.
Machine learning identifies necrosis in meningioma images
Intended benefitIIIT Bangalore presents a computational-pathology pipeline for necrosis detection, with clinical decision support as its intended application.
AI horse identification and gait analysis enter pilot testing
Intended benefitAIforPet describes real-world pilots of horse identification and gait analysis, aiming to improve animal management and welfare.
ChatZOC extends ophthalmology information and decision support
Intended benefitChatZOC pilots offer ophthalmic consultations, screening support, and teaching tools, with improved clinical access and outcomes as intended benefits.
MoralQuest proposes personalized learning for autistic children
Intended benefitMoralQuest is developing an AI-supported storytelling MVP to personalize learning and social-development activities for autistic children.
AI exercise feedback aims to support home rehabilitation
Intended benefitAnna University proposes real-time exercise assessment for rehabilitation, with laboratory and home testing planned.
The EHS website adds AI assistance and accessibility tools
Reported and intended benefitsEmirates Health Services describes a launched AI-enabled website with chatbot, multilingual, and accessibility functions; outcome gains are unspecified.
An AAC proposal aims to turn impaired speech into clear speech
Intended benefitAI4Africa proposes speech-to-speech assistance for people with communication disabilities and names Limi Hospital as a testbed.
A health-monitoring prototype targets preventive care
Intended benefitMUST proposes biosensor-based health monitoring and advice to support earlier detection and preventive care.
EquiDermAI targets skin-tone gaps in dermatology AI
Intended benefitEquiDermAI combines generated skin-lesion images and compact models to target diagnostic disparities, with full validation still ongoing.
AI journaling aims to support student mental well-being
Intended benefitStudentSphere is developing an AI journaling tool for emotional feedback, with an internal VIT pilot planned.
EEG wheelchair controls seek greater mobility independence
Intended benefitVIT reports EEG-model development using data from ten people and plans hospital testing of a brain-controlled wheelchair.
A smartwatch workflow aims to monitor facial paralysis
Intended benefitVIT describes a facial-paralysis monitoring workflow, while dataset collection and exercise-recognition model development remain planned.
AIrth proposes accessible energy and emissions insights
Intended benefitAIrth has tested an energy-insight prototype with simulated data and plans pilots with Georgian businesses and households.
AI scheduling research targets greener computing clusters
Intended benefitTeleAI describes reinforcement-learning scheduling and LLM assistance for computing clusters. Its reported GPU-availability gains come from virtualization and resource sharing.
AI sorts camera-trap imagery for biodiversity monitoring
Intended benefitSBER describes automated camera-trap analysis for 25 species, aiming to reduce manual image review and improve biodiversity monitoring.
Project Umoya plans locally adapted African weather forecasts
Intended benefitProject Umoya plans to fine-tune GraphCast and Aurora with African weather data to improve local forecasts for climate-sensitive decisions.
Weather models guide power-line communications research
Intended benefitUPM studies how regional weather affects power-line communications in Malaysia and Nigeria, aiming to support more resilient network decisions.
Generative scene reconstruction targets disaster communications
Intended benefitTCS proposes semantic scene understanding and generative rendering to help disaster responders share visual information under bandwidth constraints.
An ocean dead-zone prototype aims to provide early warnings
Intended benefitVIT is developing a system to forecast ocean hypoxia two to four weeks ahead, with regional pilots planned.
AI-assisted radio localization aims to help search and rescue
Intended benefitMUST proposes machine-learning analysis of radio signals to locate people during search-and-rescue operations.
An urban heat advisor proposes targeted cooling measures
Intended benefitChrist University is developing an urban heat advisor to identify vulnerable areas and support targeted cooling and energy decisions.
Deep learning studies local microclimate prediction
Intended benefitA microclimate research model aims to inform urban design, heat-risk management, and irrigation through more detailed local forecasts.
Proposed ML predicts elephant conflicts from movement logs
Intended benefitMUST proposes learning from elephant movement logs to predict conflicts; its existing GPS and fixed-geofence prototype is not counted as an AI benefit.
Wind-forecast AI targets energy and climate decisions
Intended benefitResearchers are developing a wind-model proof of concept for the Red Sea coast and Gulf of Suez, with operational validation still planned.
An Ethiopian chatbot proposal supports disaster assistance
Intended benefitThe Ethiopian Red Cross and meteorological partners propose a disaster chatbot, with humanitarian deployment pilots planned.
A low-code AI platform supports monitoring applications
Intended benefitChina Telecom HongXin describes an internal low-code platform for developing AI applications across transport, governance, and environmental monitoring.
A telecom model toolchain supports intelligent-agent development
Intended benefitChina Telecom describes an internal model-and-agent toolchain intended to streamline cloud-network operations and collaborative development.
Network digital twins seek lower telecom energy use
Intended benefitTata Elxsi is testing AI-driven telecom digital twins in simulated environments to identify energy-saving configurations.
AI ionospheric-delay prediction for wireless positioning
Intended benefitThe Venus project describes real-time AI prediction of ionospheric delay. A separate improvement in positioning or communication quality is not established.
AI radio sensing proposes new network and monitoring functions
Intended benefitHFCL proposes AI-based radio-frequency sensing for telecom, transport, manufacturing, and monitoring applications, without a reported testbed outcome.
Learned traffic prediction aims to improve network routing
Intended benefitIIT Gandhinagar proposes learned traffic forecasts and GNN/DRL policies to support adaptive routing, alongside conventional network optimization.
Dynamic 5G slice switching aims to improve network resilience
Intended benefitIIT Hyderabad is implementing dynamic switching of network-slice instances to respond to changing load.
Predictive beamforming aims to maintain links in bad weather
Intended benefitAddis Ababa Science and Technology University proposes weather-aware beamforming, with simulation and toolkit development planned.
Federated anomaly detection targets 5G network security
Intended benefitIIT Hyderabad is implementing federated anomaly detection using network emulators, aiming to improve 5G security and service resilience.
A GIS digital twin aims to optimize base-station energy use
Intended benefitAmrita University combines GIS data and a network digital twin to explore more adaptive and energy-efficient base-station operation.
A virtual 3D demonstrator targets remote teaching
Intended benefitTCS presents an initial proof of concept for bandwidth-efficient interaction with a virtual demonstrator, aiming to extend remote teaching and training.
Weather-aware optical communications remain in simulation
Intended benefitMUST has simulated an optical-communication design while training an AI attenuation model, with hardware integration still planned.
A bilingual AI model checks nonprofit-name similarity
Reported and intended benefitsThe UAE ministry uses fine-tuned Sentence-BERT for bilingual name similarity, alongside conventional database and prohibited-term checks.
GaP plans an AI interface for political knowledge
Intended benefitThe Centre for Gender and Politics proposes multilingual NLP, contextual recommendations, and mentorship matching for young women in South Asia.
HEART Engine uses AI to personalize training and matching
Reported and intended benefitsHEART Engine describes AI-assisted learning pathways, conversational support, and job matching; learning and employment gains are not quantified.
Radiolytica develops AI monitoring for community radio
Intended benefitFondation Hirondelle is developing radio transcription and content analysis for Eastern DRC, aiming to improve responses to harmful narratives.
A research proposal makes digital-service terms easier to read
Intended benefitA Zambian research project proposes summaries and warnings for potentially harmful terms-of-service and privacy-policy clauses.
UNHCR studies AI support for refugee-status processing
Intended benefitUNHCR is developing process analysis to identify refugee-status bottlenecks, with pilots planned and no achieved time saving reported.
FAHR adds AI assistance to federal human-resources services
Reported and intended benefitsFAHR launched an AI-enabled HR app across UAE federal entities, adding assistance and automation without published outcome measurements.
UNHCR prototypes a virtual legal assistant
Intended benefitUNHCR is developing an assistant to collect, summarize, and analyze laws affecting displaced and stateless people. Research-time savings remain a target.
AI transcribes radio broadcasts for monitoring in eastern DRC
Reported and intended benefitsFondation Hirondelle reports that AI transcribes ten radio stations in Swahili and French for content monitoring in eastern DRC. Transcription quality and downstream impact remain unmeasured.
Smartphones become AI-managed meeting microphones
Intended benefitMUST pilots a smartphone audio system with automated speaking queues and transcription. Lower equipment costs and smoother participation are expected.
An AI intrusion detector for IoT networks
Intended benefitAn edge-and-cloud intrusion-detection proposal aims to protect IoT systems. Detection performance and security benefits remain unreported.
Etcera tests AI support for demanding knowledge work
Intended benefitEtcera is testing a multi-agent assistant intended to reduce decision fatigue and cognitive overload. Productivity and well-being outcomes remain unmeasured.
Dubai proposes AI talent matching
Intended benefitDubai's talent-matching proposal aims to connect job seekers with suitable roles and reduce hiring bias. Employment outcomes are not yet reported.
Virtual try-on for Indian fashion
Intended benefitA generative-AI fashion proposal combines virtual try-on and garment design. Lower returns, less waste, and improved artisan opportunities are intended benefits.
Inspection robots for hazardous industrial environments
Intended benefitSevnce pilots an explosion-proof inspection robot. Inspection, cost, and reliability improvements remain projected.
ChatCAD brings natural-language assistance to design
Intended benefitAInnovation completed a ChatCAD MVP integrated with Bentley software. Faster design and shorter product-development cycles remain intended benefits.
GAC applies AI across automobile manufacturing
Intended benefitGAC and Huawei describe AI for design, inspection, maintenance, and scheduling. Reported time and capacity figures are presented as potential improvements.
XR training for industrial automation
Intended benefitBSD design proposes AI-supported extended-reality training to improve workers' automation skills. Learning, production, and onboarding outcomes are not reported.
AI sensors assess concrete condition
Intended benefitMUST is developing image and sensor models for concrete cracks and moisture. Field maintenance, safety, and material-saving benefits remain intended.
Zamtel pilots AI-enabled financial inclusion
Intended benefitZamtel is piloting mobile-network credit scoring and fraud detection. Improved borrowing access, incomes, and consumer protection remain intended benefits.
AI fraud screening for NGO operations
Intended benefitResearchers propose automated NGO verification, transaction monitoring, and audits. Fraud and audit-efficiency improvements are projections.
AI assistant answers UAE FATCA/CRS reporting inquiries
Reported benefitThe UAE Ministry of Finance reports that its website AI assistant gives real-time FATCA/CRS guidance. Changes in service quality or reporting work are unmeasured.
Receipt Guardian helps families practice budgeting
Intended benefitReceipt Guardian is a tested beta app for children and families. Better financial literacy and spending habits are intended, rather than measured, outcomes.
Adaptive credit thresholds tested in simulation
Intended benefitVIT researchers propose dynamic credit-scoring thresholds for subprime lending. Tests use synthetic simulations; real borrowing and lending benefits are not reported.
InsurVision plans automated insurance-claim analysis
Intended benefitIIT Hyderabad is developing InsurVision to analyze claims. Faster settlements and disaster recovery remain intended benefits, with implementation still in progress.
SERA proposes AI budgeting and financial guidance
Intended benefitSERA proposes AI models that personalize budgets, identify spending reductions, and suggest income opportunities. Changes in users’ finances are unmeasured.
Offline AI learning for universities in crisis zones
Intended benefitAl-Aqsa University describes a low-power, offline learning system for crisis settings. Learning continuity and instructor workload benefits remain expected.
An AI instructor for a virtual electronics laboratory
Intended benefitAn Ethiopian project plans a nine-month virtual-lab pilot with AI feedback and Amharic support. Access and skill gains remain prospective.
Adaptive AI learning for neurodivergent students
Intended benefitA planned special-education pilot aims to adapt teaching to students' cognitive and emotional needs. Learning and well-being effects are not yet reported.
UN Women uses generative AI in practical learning and work
Reported and intended benefitsUN Women describes hands-on use of generative AI and custom GPTs for research, knowledge products and advocacy. AI’s effect on learning or workplace performance is unmeasured.
Little Signs pilots classroom sign-language access
Intended benefitLittle Signs is piloting speech-to-sign-language support at a Zimbabwean school. Improved classroom access and participation remain unmeasured.
Personalized AI services across education and livelihoods
Intended benefitRein4ce proposes personalized AI for learning, finance, nutrition, and health. The cross-domain concept has no reported implementation benefit.
Culturalink connects cultural knowledge across countries
Intended benefitCulturalink organizes heritage information and generates cultural narratives. Learning, representation, and preservation benefits are described but not measured.
Yuanjing proposes AI-assisted urban management
Intended benefitChina Unicom describes AI dispatch, monitoring, and event management. Improvements in city services remain unmeasured.
AI-generated video ringback tones
Intended benefitiMUSIC proposes personalized video ringback tones. The project is preparatory, with no pilot or realized communication benefit reported.
Yi Jiang Jie offers AI-guided museum visits
Intended benefitChina Telecom describes an interactive museum guide with contextual explanations and personalized routes. Visitor and institutional outcomes remain unmeasured.
Proposed AI model for satellite-dish realignment
Intended benefitMUST proposes training an AI model on dish-misalignment patterns to improve realignment. Its current design uses stored angles and feedback controls; AI-related gains remain unmeasured.
Predicting network quality for online events
Intended benefitAI4Africa proposes forecasting connection quality to help plan online events. Avoided disruption and access benefits are not yet reported.
AI optimizes layouts for low-income housing
Intended benefitVIT is testing space optimization on synthetic and reference floor plans. Improved living space and construction outcomes remain prospective.
A more inclusive AI drowsiness detector
Intended benefitCatalink is validating a driver-monitoring prototype. Fewer crashes and safer public transport remain intended outcomes.
UNHCR develops Kurdish speech transcription
Intended benefitUNHCR is fine-tuning Kurdish speech recognition for humanitarian interviews. Faster processing and more inclusive decisions remain intended benefits.
A prototype for Indian Sign Language translation
Intended benefitIIIT Hyderabad is developing an ISL recognition and translation proof of concept. Communication and education benefits remain prospective.
Eye gestures for hands-free computer control
Intended benefitAn Ethiopian project combines vision models and eye tracking for computer control. Functional benefits for users remain intended while development continues.
Safer autonomous wheelchair navigation in development
Intended benefitSUPSI and IDSIA developed a laboratory demonstrator for autonomous wheelchair safety. Independent mobility benefits remain prospective.
Simulating coordinated vehicle decisions
Intended benefitTCS is researching low-cost decision making for vehicle groups. Safer traffic and lower fuel use remain intended benefits of the simulated approach.
V2V coordination for narrow and congested roads
Intended benefitOptM describes controlled pilots of vehicle communication and AI coordination. Safer journeys, smoother traffic, and lower emissions remain expected.
AI-enabled 6G vehicle communications
Intended benefitAI4Africa proposes 6G-V2X networks for connected transport. The system has no deployment, and mobility benefits remain prospective.
Detecting attacks on vehicle communications
Intended benefitAI4Africa proposes intrusion detection for V2V networks. No intercepted attack or safety improvement has been reported.
Simulating AI advice for electric-vehicle endurance
Intended benefitResearchers report simulated energy-efficiency gains of about 9.2% from driving advice. The system has not been deployed, so real vehicle and emissions benefits remain unknown.
Local soil and weather intelligence for farmers
Intended benefitA TinyML weather-monitoring prototype supports a proposed farm-advisory system. Yield, income, and resource-saving effects are not reported.
Zimbabwe pilots an agricultural voice-advice center
Intended benefitPOTRAZ describes pilots in three wards using voice and SMS agricultural advice. Income, loss-reduction, and adoption figures are projections.
Edge AI detects tomato diseases on a demonstration farm
Intended benefitMUST reports a tomato-disease detection pilot using edge AI. Avoided crop losses, higher yields, and reduced pesticide use remain unmeasured.
Soil sensors support proposed fertilizer advice
Intended benefitMUST is developing sensor-based soil analysis and a random-forest advisory model. Yield improvements and lower fertilizer waste remain intended.
Proposed AI decision support for human resources
Intended benefitA UAE research project proposes AI decision support for HR personnel. Better decisions, lower costs, and reduced hiring bias remain intended benefits.
Proposed AI rainfall prediction for basin irrigation
Intended benefitIARI proposes AI rainfall prediction to avoid unnecessary irrigation when rain is expected. The report separates this proposed extension from its existing soil-moisture automation.
Nara proposes AI voice control for an accessible metaverse
Intended benefitNara's patent specifies AI voice control for an accessible metaverse. The Blind Metaverse demo uses speech synthesis; deployment of the patented speech recognition is unverified.
Milton Keynes offered a practical AI workshop and interactive demonstrations in 2024. Better understanding, confidence, and feedback remain intended benefits without measured changes.
Dorset plans acoustic monitoring in six care homes
Intended benefitDorset plans 184 acoustic and movement sensors to support care. The source reports intended sleep and falls benefits, before outcome evaluation.
Warwickshire food-mapping AI trails simpler method: 63% versus 95%
Intended benefitWarwickshire’s LLM achieved 63% classification accuracy versus 95% for a simpler method. Faster food-access mapping remains intended; improved food access is not established.
Newham proposes AI forecasting for temporary-accommodation planning
Intended benefitNewham Council and the University of East London propose using existing data to forecast housing demand and homelessness trends. Better temporary-accommodation planning remains an intended benefit.
Microsoft's MatterGen generates TaCr2O6, measured at 169 GPa
Category not assignedMicrosoft's diffusion model generated a structure conditioned on bulk modulus 200 GPa; SIAT (Shenzhen) synthesized it and measured 169 GPa, with Ta/Cr compositional disorder. Vendor co-authored proof.
a bulk modulus of 200 GPa and measured at 169 GPa
FamilySearch and Transkribus make handwritten records searchable
Category not assignedFamilySearch introduced AI handwriting-recognition full-text search via FamilySearch Labs in 2024; Transkribus (READ-COOP) reports 200+ million pages deciphered. Makers' accounts only; no finding.
200 million pages deciphered
Liverpool plans multilingual AI concierges for summit attendees
Intended benefitLiverpool's 2024 AI Summit planned interactive digital humans to assist attendees and host proceedings. Easier access to event information remains an intended benefit without a measured outcome.
See & Spray saves 28.4–62.4% of postemergence herbicide
Category not assignedArkansas-Mississippi trials (2021–2022) with Blue River co-authors: targeted sprays matched broadcast weed control within 1% while saving 28.4% to 62.4% of postemergence herbicide.
on average saved 28.4% to 62.4% of postemergence herbicide
ChatGPT helps review MRI literature on intravascular lymphoma
Category not assignedChinese neurologists used ChatGPT to help review published MRI findings of cerebral intravascular large B-cell lymphoma for a 59-year-old's case report; the diagnosis came from biopsy, not the model.
KoBold used AI to characterize the Mingomba copper deposit
Reported benefitKoBold combined AI with geological expertise to characterize Mingomba; the sources do not isolate AI’s effect on discovery, costs, or environmental impact.
Project Amber fills 5,145 Blackpool potholes for about £450,000
Category not assignedBlackpool's Project Amber imagery/AI surveys: 5,145 potholes filled for ~£450k; pothole claim payouts fell to £719 (DfT, 2024)
5,145 potholes were filled in North Shore for just under £450,000
Austin Energy deploys 13 Pano AI cameras across its territory
Category not assignedThe City of Austin's utility announced full deployment of a 13-camera Pano AI detection system covering its service area with the Austin Fire Department (29 August 2024); no detections are reported.
Waymo has zero bodily-injury, 76% fewer property-damage claims
Category not assignedHeliyon (14 July 2024, Swiss Re and Waymo authors): Waymo liability claims benchmarked against 600,000+ Swiss Re claims; a Dec 2024 25.3M-mile update reports 88%/92% reductions.
0.78 property-damage claims per million miles against 3.26, a 76% reduction
XR Guild proposes an AI librarian for ethics resources
Intended benefitXR Guild proposes an AI librarian to find ethics resources; public catalog pages also show ChatGPT-attributed keywords, but improved discovery remains unmeasured.
ENEA tests ML heatwave grid-failure forecasts; no outcome given
Category not assignedThe OECD AI Incidents Monitor records that ENEA, Politecnico di Bari and Università Roma Tre developed and tested an ML system to predict heatwave-induced grid failures (May 2023); no outcome given.
Nelli flags 247 nocturnal seizure events missed by carers
Category not assignedCornwall NHS evaluation (15 people who could not tolerate EEG): Nelli flagged 707 possible seizure events, 247 unheard or unrecognised by carers; Danish phase-3 validation sensitivity 93.7%.
Nelli flagged 707 possible seizure events, 247 unheard or unrecognised by carers
Germany tests IdentiFlight cameras for bird protection
Category not assignedAn OECD entry reports German deployment in September 2022. OekoFor's undated firsthand account describes German test-site validation, without a measured reduction in bird fatalities.
Confirmed TREWS sepsis alerts are linked to 3.3% lower mortality
Category not assignedProspective five-hospital study: sepsis alerts confirmed within 3 h had 3.3% absolute lower in-hospital mortality (CI 1.7–5.1).
3.3% absolute (18.7% relative) reduction in in-hospital mortality
ABAIR's Irish voices power an NVDA screenreader for blind pupils
Category not assignedSLPAT 2022 paper by the ABAIR team (Trinity College Dublin): Irish NVDA plugin with Braille output, NCBI-funded, built with a blind researcher, tested with visually impaired pupils; no dated outcome.
EasyScan GO reaches 91.1% sensitivity for malaria at 11 sites
Category not assignedDas et al. 2022: 2,250 blood films at 11 sites in 11 countries (2018–2019); EasyScan GO sensitivity 91.1%, specificity 75.6% vs expert microscopy; no patient outcome; Global Good funded.
Diagnostic sensitivity was 91.1% (95% CI 88.9–92.7) and specificity 75.6% (73.1–78.0)
Faculty's Kettering bed tool is open-sourced but not deployed
Category not assigned14-week NHS AI Skunkworks/Faculty proof of concept (2021) forecasting demand and suggesting beds at Kettering General Hospital; tested by the hospital, open-sourced, not in live use.
OCHA uses 510's model as CERF trigger; no activation documented
Category not assignedOCHA frameworks (2021, 2023, 2025) use 510's XGBoost house-damage model as activation trigger; near-activation Dec 2021 (Rai); the Nov 2025 activation for Uwan ran on the 185 km/h wind trigger.
icipe MaxEnt model maps locust breeding, mean AUC up to 0.887
Category not assignedPeer-reviewed MaxEnt model trained on 5,406 FAO breeding records (1985–2020) projected breeding suitability for Kenya, Uganda, South Sudan and Sudan; no operational outcome reported.
the Mauritania model had the highest mean AUC (0.887)
More loan approvals at lower rates in a simulation
Category not assignedUpstart estimated 27% more approvals and 16% lower annual percentage rates in a simulation. A 2024 independent review found approval disparities for Black applicants.
approving 27% more applicants at 16% lower APRs than a traditional model
HumCat proposes AI support for catastrophe prevention
Intended benefitKYield proposes HumCat for organizational risk prevention, but the reviewed sources do not document a catastrophe that it prevented.
Karolinska eye-tracking SVM flags reading risk at 95.6% accuracy
Category not assignedNilsson Benfatto et al. trained an SVM on under-one-minute eye-tracking records from 97 high-risk and 88 low-risk Swedish children; 95.6% accuracy; two authors own equity in Optolexia.
reached 95.6% (± 4.5%) accuracy with balanced sensitivity and specificity
Corti model beats dispatchers on sensitivity; RCT shows no gain
Category not assignedCopenhagen EMS call model beat dispatchers on sensitivity (84.1% vs 72.5%); RCT found no significant gain in dispatcher recognition.
higher sensitivity than dispatchers (84.1% versus 72.5%)
Superpower Glass raises Vineland socialization by 4.58 points
Category not assignedStanford's Superpower Glass cut SRS-2 scores 7.38 points in 14 children and raised Vineland socialization 4.58 points in a 71-child RCT.
gained 4.58 points more on the Vineland socialization subscale than controls (P = .005)
CapisciAMe CNN outperforms familiar listeners on impaired speech
Category not assignedA CNN trained only on one patient's voice recognized 72.65% of her words versus 56.75% for 12 familiar professionals; lab result only.
recognized 72.65% of her words versus 56.75% for 12 familiar professionals
CAD4TB triage of 94,694 people in Nigeria yields 2,337 TB cases
Category not assignedKNCV/USAID deployment Jan–Sep 2022: CAD4TB v7 screened 94,694 people in eight Nigerian states; 2,337 TB cases diagnosed, 97% treated
2,337 TB cases diagnosed, 97% treated
CMS's FPS identifies or prevents $654.8 million in 2015
Category not assignedCMS's return-on-investment report for the Fraud Prevention System's fourth implementation year counts $654.8 million in inappropriate Medicare payments identified or prevented in 2015 against $57.1...
$654.8 million in inappropriate Medicare payments identified or prevented in 2015
AI teams out-earn growers in Wageningen's greenhouse challenges
Category not assigned2018 cucumber challenge: winning algorithm raised sellable yield 6% and net profit 17% over reference growers; 2020 cherry-tomato challenge: all five AI compartments beat the growers' net profit.
winning algorithm raised sellable yield 6% and net profit 17% over reference growers
Microsoft Copilot tutoring lifts Benin City scores by 0.31 SD
Category not assignedWorld Bank Policy Research Working Paper 11125: teacher-guided after-school Copilot (GPT-4) sessions, June-July 2024, Benin City; 0.31 SD overall, 0.23 SD in English; 657 treated, 671 control.
0.31 SD overall, 0.23 SD in English
Rising Academies' WhatsApp tutor Rori lifts math scores 0.36 SD
Category not assignedPreprint by Oxford, Rising Academies and J-PAL North America authors: students using Rori one hour a week in study hall gained 0.36 SD more in math (637 baseline, 477 endline) than controls in 2023.
an effect size of 0.36 (Cohen's d, p < 0.001)
Camera AI on Spin e-scooters cuts time ridden on sidewalks 22%
Category not assignedUniversity of Washington study in Findings (20 Nov 2024): feedback disabled at random on 50 of 100 camera-equipped Spin scooters from 23 Nov 2022 to 14 Feb 2023; feedback-equipped scooters spent 22...
The feedback group spent 22% less time and 20% less distance on sidewalks
GPT-5 proposes oncocytoma first in two retrospective cases
Category not assignedSassari ophthalmologists gave ChatGPT (GPT-5) slit-lamp photos and brief summaries of two surgically confirmed caruncular oncocytomas; it proposed oncocytoma first in both. Retrospective proof only.
CMD-1 cuts crisis triage from 9 hours to 8–13 minutes
Category not assignedTwo-stage NLP (keyword filter plus logistic regression) screened 102,471 chat messages in October 2022 (AUC 0.98, sensitivity 0.98); daily median receipt-to-triage time 8–13 min versus 9 h before.
daily median receipt-to-triage time 8–13 min versus 9 h before
GFW analysts help Indonesia seize MV NIKA; no AI detection
Category not assignedMV NIKA seized by Indonesia 12 July 2019 after INTERPOL tracking with OceanMind and GFW; captain convicted, owner fined, vessel forfeited; no cited source documents an AI detection.
the captain was convicted, the owner fined Rp300,000,000 and the vessel forfeited
RapiD AI-assisted mapping up to 12.4% slower, then disabled
Category not assignedOSM wiki: University of Colorado Boulder/Indiana University research found RapiD AI-assisted contributors up to 12.4% slower; RapiD disabled; OSM data used by AKUT and AFAD. No AI benefit documented.
contributors using RapiD mapped tasks up to 12.4% slower
AIMS's ReefScan and ReefCloud automate reef image analysis
Category not assignedAIMS states ReefScan uses machine learning for real-time object classification and ReefCloud automates benthic image analysis; no reef management decision attributed to the AI was found.
Whale Safe grades ship cooperation with 10-knot slowdowns
Category not assignedWhale Safe rates whale presence from acoustic buoys, sightings and a habitat model and grades ship cooperation with California's voluntary 10-knot zones; no avoided strike is documented.
GrantRail demonstrates AI support for grant decisions
Intended benefitGrantRail demonstrates intended AI support for education-grant decisions using synthetic data; its displayed portfolio figures are not real outcomes.
AI copy appears among campaign ads with high click rates
Reported benefitBattlegroundAI reports that its AI copy appeared among school-board campaigns’ strongest ads, without establishing an effect on turnout or election results.
A proposal links records to support people facing homelessness
Intended benefitA Darkriderdesign submission proposes Copilot-assisted record synthesis for social workers; this review found no external account of implementation or results.
A retailer predicts style preferences from consented data
Reported benefitOpen Data Labs says that an unnamed retailer improved fashion-preference prediction using consented data; the provider supplies no accuracy values or independent evaluation.
StratLoom uses AI within instructional workflows
Intended benefitStratLoom describes AI-assisted curriculum work and supervised prompt practice; its public sources do not establish improvements in learning or educator efficiency.
No cases match.