Adyt

AI system

Apple Watch Fall Detection

Developer

Apple

AI contribution

The paper reports that the watch recognized the fall and automatically called 112 and alerted emergency contacts. Those alerts brought responders to the man behind the hedge. Humans provided resuscitation, defibrillation, and surgery. The watch's contribution is documented in the clinical narrative, but the time to rescue or outcome without it cannot be measured from this case. Apple's documentation describes the detection and notification sequence without disclosing the classifier or model version.

Comparison

being found without the automatic alert, on a remote spot behind a hedge

Before

Zonder die melding had ik hem niet zo snel gevonden (without that alert I would not have found him so quickly)

After

CPR approximately 6–7 min after collapse; full neurological recovery

Comparison basis

Assumed

Full case and sources

AI system

Udbetaling Danmark / ATP Den Faelles Dataenhed fraud-control models (about 60 profiling and machine-learning models as of 2019, including 'Really Single', 'Model Abroad' and 'Fictitious Employment')

Developer

ATP (Arbejdsmarkedets Tillaegspension) with NNIT A/S as supplier, for Udbetaling Danmark

Deployer

Udbetaling Danmark (public authority), administered by ATP

AI contribution

The models select “requested cases” that UDK's HOK control team screens and investigates. Decisions to stop or reduce benefits are taken by humans after investigation. UDK measures model quality by the share of cases the team accepts for manual processing. Every outcome figure is a UDK statement (FOI responses and a letter) as reproduced verbatim in Amnesty International's report. No official publication of them is among the sources. Amnesty contests effectiveness and alleges discrimination risks, and UDK rejected parts of Amnesty's assessment. NNIT built the models to ATP's specifications. No developer-authored source is on record. The dm.dk report and ATP's page are not among the sources.

Full case and sources

AI system

'Ghost Murmur': long-range quantum (nitrogen-vacancy diamond) magnetometry paired with AI software that isolates heartbeat signatures from background noise, as described by the NY Post's anonymous sources

Developer

Lockheed Martin Skunk Works, per two NY Post sources; the company declined to comment

Deployer

US Central Intelligence Agency, per the NY Post; the CIA and White House named no tool

AI contribution

The Post's anonymous sources describe AI that isolates a heartbeat signal from sensor noise. No public technical evidence establishes that this system located the airman. The official account confirms the rescue and camera observation, while the later presidential remark does not establish the sensing method.

Full case and sources

AI system

Lynxight AI pool monitoring with CCTV analysis and smartwatch alerts

Developer

Lynxight

Deployer

City of Stirling

AI contribution

Lynxight detected distress and alerted the lifeguard, who had not seen the man sink. Human rescuers brought him to safety. Whether he would have survived without the alert is unknown.

Full case and sources

AI system

Viz.ai stroke detection and care coordination (attribution inferred)

Developer

Viz.ai (inferred)

Deployer

Baptist Health, Marcus Neuroscience Institute

AI contribution

AI flagged the bleeding and notified the stroke team, according to her treating surgeon. Clinicians treated the aneurysm. Viz.ai is an inferred identification from his separate interview. The sources do not isolate the software’s effect on survival.

Full case and sources

AI system

Meta proactive self-harm detection on Instagram with alert routing to the Uttar Pradesh Police Social Media Center

Developer

Meta Platforms

Deployer

Meta Platforms with the Uttar Pradesh Police Social Media Center (agreement signed March 2022)

AI contribution

Police sources say Meta sent the alert after the post. They do not state whether Meta's classifier or a user report flagged it. Meta's documentation and NPR describe the process as machine-learning detection with human review before contacting authorities. The record relies on that description and should be read with that caveat.

Full case and sources

AI system

Meta's suicide and self-injury detection and review pipeline on Instagram, with email alerts to the Uttar Pradesh Police Social Media Center at DGP headquarters

Developer

Meta Platforms

Deployer

Meta Platforms with the Uttar Pradesh Police Social Media Center (Decode: Uttar Pradesh has been part of Meta's suicide-prevention program since 2022)

AI contribution

Police say Meta flagged the Instagram post at about 8:58 pm on 29 March and emailed the Uttar Pradesh Police Social Media Center. No source says whether Meta's classifier or a user report started the review. Decode reports that UP alerts arise from “artificial intelligence systems or user reports” reviewed by Meta staff. Everything after the email was human work. The center traced the phone number and location. Raebareli police reached the home in 12 minutes on the DGP's instructions, and a hospital treated him.

Comparison

reports escalated without Meta's pattern-recognition acceleration

Before

escalation speed of other reports

After

escalated to local authorities twice as quickly as other reports

Comparison basis

Measured

Full case and sources

AI system

Vyntra (NetGuardians) AI risk models for real-time payment and e-banking fraud detection

Developer

Vyntra (NetGuardians SA, merged with Intix in June 2025)

Deployer

Unnamed client banks

AI contribution

The evidence is limited to the maker's account. Vyntra's page attributes each block to its AI risk models detecting variables outside the customer's or bank's usual profile. No customer, bank, date or official record is available to assess whether the models, rather than staff or other controls, produced the outcome. The AI's contribution is not established.

Full case and sources

AI system

ASENA (Analiz Sistemleri Narkotik Ağı), Turkish National Police narcotics analysis software

Developer

Turkish National Police (Emniyet Genel Müdürlüğü), Narcotics Crimes Department; described as written by Turkish engineers, no contractor named

Deployer

Hakkari Provincial Police Directorate, Turkish National Police

AI contribution

The program supplied the lead and nothing more. The minister said the analysis programs, “what they call artificial intelligence”, had identified a suspicious vehicle thought to be carrying drugs. The Hakkari police chief called him on the evening of 20 March. Officers then stopped the car at a fixed checkpoint, searched the concealed compartment and arrested the driver; prosecutors and the court did the rest. No source describes the program's inputs, the data it matched, or whether an analyst reviewed the flag before the stop. The defendant spoke in court of “those who informed”, but no source reports an informant.

Full case and sources

AI system

WeChat Pay 'anti-fraud customer service reminder' (防骗客服提醒) with Tencent Cloud Zhifu (腾讯云智服) online customer service

Developer

Tencent (WeChat Pay and Tencent Cloud)

Deployer

WeChat Pay

AI contribution

The automated part the sources describe is the flagging of a risky transaction. That triggers the on-screen reminder and connects the payer to Tencent Cloud Zhifu, a service the Monitor calls AI. ChinaZ says the WeChat Pay customer service judged “from experience” that this was a dating scam, which reads as a human agent. Neither report says how the transaction was flagged or whether any model was involved. The advice, not the flag alone, led Zhang to cancel.

Full case and sources

AI system

SARUAV (Search and Rescue with Unmanned Aerial Vehicle) human detection software, SARUAV, run on images from DJI Matrice 300 drones

Developer

SARUAV Ltd. / University of Wrocław (team of Tomasz Niedzielski)

Deployer

Bieszczady Mountain Rescue Service (GOPR Bieszczady)

AI contribution

The CNN detector returned the man's coordinates after police and tracking dogs had failed. Rescuers then retrieved him. Three of the five case-report authors work for SARUAV Ltd., so the independent confirmation is the GOPR Bieszczady account quoted in the Polish press.

Comparison

the ground search by police officers and rescuers with tracking dogs, which had failed

Before

Psy zgubiły trop (the dogs lost the scent); police and rescuers could not find him

After

Po prawie 24 godzinach od zginięcia program wskazał potencjalnego poszukiwanego (after almost 24 hours the program indicated the probable missing person)

Comparison basis

Measured

Full case and sources

AI system

Fire and Rescue NSW remotely piloted aircraft system with thermal imaging and AI person-detection software (software vendor not named in the sources)

Deployer

Fire and Rescue NSW

AI contribution

The AI flagged the hikers in the drone feed, and the thermal camera confirmed the find. The inspector overseeing the search credits it with cutting the search short. The hikers also signaled with a phone light, and the software vendor is not named. FRNSW's own media release was not available and is not cited. No cited source dates it.

Comparison

a conventional search without the AI drone

Before

several days (the inspector's estimate of the search time avoided)

After

less than five hours after they were reported missing

Comparison basis

Assumed

Full case and sources

AI system

TrailGuard AI camera network (about 220 units) with embedded human/animal/vehicle classification

Deployer

Similipal Tiger Reserve, Odisha Forest Department (Similipal South Division, Special Tiger Protection Force, Similipal Wildlife Intelligence Network)

AI contribution

The AI cameras' real-time detection of the armed group is the stated trigger for the mobilization. That operation ended in surrender and seizure. The park authority's statement is the Field Director's on-record account, quoted by The New Indian Express and the Times of India. The court remands are reported in the press by PTI, Hans India and Ommcom. The cited account lacks corroboration from an available official press note or court order. The account rests on press-quoted forest-department statements.

Full case and sources

AI system

thynC (씽크) AI-based smart bed monitoring system: wearable sensors measuring ECG, heart rate and respiratory rate in real time, with alarms sent to a remote central monitor

Developer

Not stated in the read reports, which describe thynC as the smart bed monitoring system that Daewoong Pharmaceutical supplies (자사 공급)

Deployer

Gwangju Hanguk Hospital (광주한국병원)

AI contribution

The system's contribution was detection and summons. The wearable ECG stream showed the heart rate climbing to 192 beats per minute. The remote central monitor identified ventricular tachycardia and raised an alarm. The attending director says that is how staff learned of it at once and reached the bedside to begin CPR within the golden time. The resuscitation itself was done by people. The counterfactual rests on the hospital's framing that general-ward patients are not under continuous observation and that deterioration can be noticed late. How long this patient would otherwise have gone unobserved is not recorded. No medical record or independent assessment has been published. The case reached the press through a Daewoong Pharmaceutical announcement of 11 February 2026.

Full case and sources

AI system

ChatGPT, Not named; Dani said he used 'el modelo más inteligente de ChatGPT' (the most intelligent ChatGPT model)

Developer

OpenAI

AI contribution

ChatGPT produced the insulinoma hypothesis from a history that two doctors had reviewed without reaching it. That prompted the glucose sensor and the emergency visit. The diagnosis was made by an internist who, Dani wrote, “lo identificó sin nosotros decirle lo que creíamos que era”. The surgery and recovery were the hospital's work. Dani says he is not a doctor but has “formación biosanitaria de muchos años”, which let him understand what he was doing. Every account traces to his own video.

Comparison

three years of visits to specialists, including a neurologist, and a primary-care doctor

Before

Tras varias pruebas le “dijeron que no era nada”

After

en pocos minutos su diagnóstico señalaba como posible principal causa de lo que me sucedía: un tumor de páncreas, un insulinoma

Comparison basis

Assumed

Full case and sources

AI system

U.S. Customs and Border Protection (Office of Field Operations) machine-learning model that scores vehicle border-crossing histories for suspicious patterns; not named in any source

Deployer

U.S. Customs and Border Protection, Office of Field Operations (U.S. Department of Homeland Security)

AI contribution

The model's contribution was the referral. DHS says it identified a suspicious pattern in the car's border-crossing history and flagged it for review. Mayorkas said the anomaly was identified in 1.4 seconds. The discovery of the narcotics and the arrest were done by officers in secondary review. No source states whether the car would have been referred without the flag. The model's counterfactual contribution is thus asserted by DHS, not demonstrated. The Monitor entry also bundles Mayorkas's claim that AI age-progression imagery has recovered abducted children. That claim names no case, date, or child.

Full case and sources

AI system

運動心電圖AI輔助判讀系統 (Exercise ECG AI-Assisted Interpretation System), China Medical University Hospital, Not reported. The hospital describes a deep-learning model that takes exercise-ECG signals at three time points plus 10 measured and 2 derived features (such as history and maximum workload), trained on nearly 1,000 exercise ECGs matched to catheterization data, accuracy about 80%, sensitivity 88%, in clinical use since May 2022; the authors' 2025 paper describes a convolutional recurrent network with three ECG slices and 12 features, AUC 0.83, sensitivity 0.89, specificity 0.60.

Developer

China Medical University Hospital (中國醫藥大學附設醫院人工智慧中心), director 許凱程

Deployer

China Medical University Hospital, Division of Cardiology (心臟血管系), Taichung

AI contribution

The model read the exercise-ECG signals and report features. It returned a 72% probability of severe coronary narrowing, above the moderate-risk verdict of the manual reading and risk score. On that basis the cardiologist ordered angiography, which showed the 99% stenosis, and he stented it. The counterfactual rests on the treating cardiologist's statement, carried by three outlets and the hospital's release. He said the manual reading would have led to drug treatment and follow-up only. The hospital that says so also built the model. The dates of the reading and procedure are not reported. The published model's specificity of 0.60 means such flags are often false positives.

Full case and sources

AI system

eLocust3m (PlantVillage) with AI data validation feeding FAO's Desert Locust Information Service, eLocust3M

Developer

PlantVillage, Pennsylvania State University, with FAO

Deployer

FAO Desert Locust Information Service and national locust control operations in Kenya, Ethiopia and Somalia

AI contribution

Two AI uses are documented: algorithms that double-check submitted photos (Times) and AI that filters unusable records out of FAO's daily intake (FAO). A Penn State deep-learning soil-moisture model is mentioned by the maker's university. The outcome figures belong to the entire control operation and no source isolates the AI's contribution. FAO is both co-developer and deployer, so its feature is maker-authored. The Times report is independent with named sources (Cressman, Ferrand, Hughes, Jeptoo, Owuor, Craig).

Comparison

no locust control operations in East Africa (FAO's estimate of the losses that the control campaign averted)

Before

loss of 4 million tonnes of cereal and 800 million litres of milk production, USD 1.56 billion in cereal and milk losses

After

those losses prevented and the food security of 36.6 million people protected

Comparison basis

Assumed

Full case and sources

AI system

SkinVision skin-check app (AI analysis of smartphone photographs of moles, graded low, medium or high risk)

Developer

SkinVision

AI contribution

The app's classifier graded a phone photograph of the mole high risk and told him to see a doctor at once. By his account that prompt, not the color change, is what sent him to his GP. He said he only goes to the doctor when absolutely desperate. The referral, excision, biopsy and staging were done by the GP, the dermatologist and pathology, none of whom the Daily Mail quotes. The article does not report the app's score, its version, or whether the 48-hour dermatologist advice it mentions was used.

Full case and sources

AI system

Karnataka Examinations Authority QR code-based facial recognition app (CET-2025 candidate verification)

Developer

Karnataka Examinations Authority (The Hindu and ETV Bharat: 'developed by KEA'; no vendor named)

Deployer

Karnataka Examinations Authority

AI contribution

The app did the detecting. Staff scanned the admit card's QR code and captured the woman's photo. It flagged a mismatch with the photo on record (New Indian Express). It “revealed that she was not the genuine candidate” (The Hindu). The Hindu and the Times of India both report the forged card's contents. It carried her own photograph with the genuine candidate's hall ticket number and QR code. Comparing the card with her face would therefore not have exposed her. The flag depended on the app's match against the registered photograph. Staff then acted, but she fled before the principal was told. No arrest or sanction is attributable to the system. KEA's probe and the promised legal action were still pending in every report.

Full case and sources

AI system

ALICE (Analisador de Licitações, Contratos e Editais), CGU text-mining and AI alert tool

Developer

Controladoria-Geral da União (CGU), in-house

Deployer

Controladoria-Geral da União (CGU)

AI contribution

The CGU release states that each tender “teve alerta do Robô ALICE” and that auditors then confirmed the overpricing risk. The suspension, revocation and re-estimation are described as results of that audit action. Human auditors and hospital managers made the decisions. ALICE's role was detection and triage. Developer and deployer are the same public body, so this is the body documenting its own tool. The CGU page is on record as an archived copy, because the live page was restricted during Brazil's 2026 electoral period.

Comparison

the hospital's original price estimate for tender 90017/2024

Before

valor estimado em R$ 15.450.714,80

After

R$ 12.386.939,96

Comparison basis

Measured

Full case and sources

AI system

Pano AI wildfire detection camera

Developer

Pano AI

Deployer

Xcel Energy (camera funding); alerts received by Red White and Blue Fire

AI contribution

The camera's AI smoke alert was the trigger for dispatch, according to the fire division quoted by CBS Colorado. The only source is a single local news report with named sources. The fire's date, cause of containment and any official report are not documented.

Comparison

locating a reported wildfire without the camera's coordinates

Before

the process of finding, which often time with wildfires can take an hour

After

get a location, an actual lat/long and drive straight through it

Comparison basis

Assumed

Full case and sources

AI system

ALERTCalifornia camera network with DigitalPath AI smoke detection

Developer

DigitalPath (AI smoke detection); ALERTCalifornia, University of California San Diego (camera network)

Deployer

California Department of Forestry and Fire Protection (Cal Fire)

AI contribution

The AI alert was the only report of the Auburn fire. Cal Fire's command center verified it and dispatched on it. The same source says the system still produces false positives from dust and clouds, so human verification sits between alert and dispatch.

Comparison

waiting for a 911 call from the public

Before

Nobody had called 911

After

put it out before it spread beyond a 30 x 30 foot area

Comparison basis

Measured

Full case and sources

AI system

CBP facial comparison biometric system (biometric entry/exit, Traveler Verification Service)

Developer

Not named in the sources

Deployer

U.S. Customs and Border Protection, Office of Field Operations, Washington Dulles International Airport

AI contribution

The facial comparison match (less than two seconds per CBP) produced the mismatch that triggered secondary inspection. The search and removal were human actions. The documentation is CBP's own release (deployer, not developer) plus NBC with a named CBP official. The algorithm vendor and the match threshold are not in the sources. CBP's “99% match rate” claim is unaudited here.

Full case and sources

AI system

Alice (Análise de Licitações e Editais), TCU/CGU text-analysis and data cross-checking robot

Developer

Tribunal de Contas da União (TCU) with the Ministério da Transparência e Controladoria-Geral da União (CGU), in-house

Deployer

Tribunal de Contas da União (TCU), Goiás unit

AI contribution

Alice's role was detection. Per Inforchannel it pulls the estimated value from notice text, runs nine text analyses for competitiveness restrictions and 23 database cross-checks. The auditor's statement and the JOTA article attribute the discovery of the BR-153 notices to Alice. The cancellation was a decision of the TCU plenary (Acórdão 1116/2017). Both sources rest on TCU staff statements, the deployer describing its own in-house tool. The source counts as independent because TCU is a public body, not a vendor. The acórdão text was not obtained, so the two accounts' differing figures (R$39 million, two notices; R$20 million, one notice) cannot be reconciled.

Comparison

TCU auditors reading the roughly 200 federal tender notices published each day by hand

Before

seria muito custoso porque são, em média, 200 editais por dia

After

Os auditores do Tribunal de Contas da União recebem pontualmente às 19h um e-mail de Alice. São os resumos das centenas de contratações federais publicadas naquele dia.

Comparison basis

Assumed

Full case and sources

AI system

Iron Beam ('Or Eitan') 100 kW-class high-energy laser air-defense system by Rafael Advanced Defense Systems, with beam director and electro-optical targeting; the Monitor calls it 'AI-enabled', which no cited source supports

Developer

Rafael Advanced Defense Systems with the Israel Ministry of Defense (Lockheed Martin collaboration for the US market, per Wikipedia)

Deployer

Israeli Air Force air-defense array; first operational system delivered 28 December 2025

AI contribution

No cited source describes Iron Beam as an AI or machine-learning system. Those sources are Ynet, JPost, Army Recognition, TimesNow, The Conversation and Moneycontrol. They also include Times of Israel 28 December 2025, 16 Mar, 27 Mar and 30 June 2026, and Wikipedia. They describe a high-power laser with a beam director and electro-optical targeting. The Monitor's “AI-enabled” label is its own. Rafael's CEO described machine learning as a future capability (Breaking Defense, 22 October 2025). The IDF has not confirmed the engagement itself, so neither an AI role nor a contribution can be assessed.

Comparison

Iron Dome interceptor missiles

Before

A single Iron Dome interceptor missile costs about US$50,000

After

each shot cost around $US3.50, and more recent estimates suggest the cost may now have fallen as low as US$2.50 per shot

Comparison basis

Assumed

Full case and sources

AI system

ChatGPT (OpenAI), described in the post as 'an advanced AI model and a limited-use trial version of GPT-5', GPT-5 trial version, per the cousin's post; unverified

Developer

OpenAI

Deployer

Islam Adel, the patient's cousin, as a consumer user

AI contribution

ChatGPT's output (advice to give paraffin oil, then nine bottles in turn with breathing exercises) is documented only by the cousin's own post. The chat transcripts were not published (“too many to publish all”, per Masrawy's summary of the post). Dean Alaa Attia contradicts its effect: he says that the patient had no clinical signs of phosphine poisoning and denied swallowing the pill. Dr Nermin Hamdi's report that Assiut University denied the case involved the grain pill also contradicts it. Dr Mahmoud Amr says that oil is a recognized first aid but not a cure. Whether the hospital followed any of the AI's instructions is unsupported by any source other than the post.

Full case and sources

AI system

X-Guard fiber-optic towed decoy (Rafael Advanced Defense Systems), described by Indian outlets as 'AI-powered' with 'AI-enhanced algorithms' adjusting its emissions; no read manufacturer or official source describes it

Developer

Rafael Advanced Defense Systems (Israel); several outlets misname it 'Rafale Advanced Defense Systems'

Deployer

Indian Air Force (claimed by commentary; unconfirmed by any official source)

AI contribution

No source independent of the commentary chain describes X-Guard as an AI system. idrw writes “AI-enhanced jamming decoy” and “AI-enhanced algorithms dynamically adjust signals” without attribution. ET, India TV, LatestLY, Oneindia and TFIPOST repeat the label. The Week gives no source for “AI-enabled”. Rafael's own page was unreadable. Deployment itself is unconfirmed by the IAF, MoD, Dassault or Rafael. The no-loss outcome is contradicted by the CDS. Neither the system's AI character nor any contribution to the engagements can therefore be assessed.

Full case and sources

AI system

AI analysis of brain CT images

Deployer

Mayo Clinic

AI contribution

AI helped locate the aneurysm. Clinicians provided treatment. The software and its separate effect on survival are unspecified.

Full case and sources

AI system

DOJ Health Care Fraud Unit proactive billing analytics

Developer

U.S. Department of Justice, Criminal Division Fraud Section, Health Care Fraud Unit Data Analytics Team

Deployer

U.S. Department of Justice with HHS-OIG and CMS

AI contribution

DOJ attributes detection to proactive data analytics. The reviewed sources do not establish that AI was used. The AI-enabled Data Fusion Center was announced separately as future work.

Comparison

the amount Medicare was scheduled to pay on the fraudulent claims had the billing gone undetected

Before

the approximately $4.45 billion that was scheduled to be paid by Medicare

After

all but approximately $41 million

Comparison basis

Measured

Full case and sources

AI system

智抗菌平台 / Intelligent Anti-Microbial System (i.A.M.S)

Developer

China Medical University Hospital, Taichung (team of superintendent Der-Yang Cho 周德陽, Po-Ren Hsueh 薛博仁 of the Department of Laboratory Medicine, and Jiaxin Yu 游家鑫)

Deployer

China Medical University Hospital, Taichung

AI contribution

The platform analyzed a bacterial mass spectrum to predict resistance and suggest an alternative antibiotic. Clinicians selected and administered treatment. The hospital attributes earlier selection to the system, but the available account cannot isolate its effect on survival.

Comparison

The two to three days that hospital clinicians said conventional susceptibility testing required

Before

Two to three days

After

A prediction within one hour

Comparison basis

Assumed

Full case and sources

AI system

Unnamed TIAA AI tool used to flag suspected fraud for human review

Deployer

TIAA (Teachers Insurance and Annuity Association of America)

AI contribution

Duckett attributes the initial flag to AI and the successful intervention to a portfolio manager, the fraud team, and the participant’s daughter. The transcript does not establish the model’s design, the requested transfer amount, or what would have happened without its flag.

Full case and sources

AI system

Viz ICH Plus

Developer

Viz.ai

Deployer

Large urban healthcare system, not named in the report; the city is not stated, and all authors are affiliated with Neurosurgery, Icahn School of Medicine at Mount Sinai, New York

AI contribution

The software detected the haematoma and auto-alerted the receiving team, which coordinated stabilization advice and transfer. A CT ordered for severe headache would have shown the bleed regardless, so the AI's contribution is the speed of notification and transfer. Dr Kellner discloses research grant support from Viz.AI among other companies. No author is a Viz.ai employee.

Comparison

the average interhospital transfer time for similar intracerebral hemorrhage cases, taken from a published New York City hospital-system cohort

Before

199.7 minutes

After

101 minutes

Comparison basis

Measured

Full case and sources

AI system

ALICE (Analisador de Licitações, Contratos e Editais), CGU automated audit-trail and AI alert tool

Developer

Controladoria-Geral da União (CGU), in-house (launched 2015 per the release)

Deployer

Controladoria-Geral da União (CGU)

AI contribution

The CGU states the audits were done “com apoio da ferramenta automatizada ALICE”. At INCA, it says, the overpricing risk was identified “com a ajuda do ALICE”. For the Ipanema tender, suspension followed a CGU recommendation after analysis of the price survey. ALICE's documented role is alerting and supporting the audit; auditors and managers decided. Developer and deployer are the same public body. The CGU page is on record as an archived copy, because the live page was restricted during Brazil's 2026 electoral period.

Comparison

the original estimated value of Pregão 15/2021, which reused the price survey of Pregão 05/2021

Before

valor estimado de R$ 34 mi

After

reduziu o valor total do pregão em R$ 2.381.663,00; all items bought for R$ 14.315.793,42 on 21 July 2021

Comparison basis

Measured

Full case and sources

AI system

ALERTCalifornia camera network with AI smoke detection (the Microsoft feature does not name the detection software; AIBD-0007 identifies DigitalPath)

Developer

ALERTCalifornia, University of California San Diego (camera network); AI detection software supplier not named in the sources

Deployer

Kern County Fire Department (incident agency); which agency received the AI alert is not stated

AI contribution

The only statement that AI detection played a role is a Microsoft-published feature (Microsoft funds ALERTCalifornia) quoting its founder Neal Driscoll. His “nearly 4,000 acres” counterfactual is his estimate. The quoted deputy fire chief is also ALERTCalifornia's deputy director of operations. Cal Fire's incident record and KBAK's same-day report say nothing about how the fire was detected. KBAK attributes the start to an electrical strike.

Comparison

the fire's estimated spread without AI-powered imaging, as estimated by ALERTCalifornia founder Neal Driscoll

Before

nearly 4,000 acres

After

52 acres

Comparison basis

Assumed

Full case and sources

AI system

Empatica Embrace wrist-worn seizure detection (electrodermal activity and accelerometer, on-board algorithm) with the Alert smartphone app, Embrace (the article does not state the generation); echoes concern Embrace2

Developer

Empatica

AI contribution

Per IEEE Spectrum, the Embrace detected the seizure and its app alerted McGrath's mother, who made the emergency call. The device did not call for help itself. The article calls the algorithm proprietary. The FDA summary describes it as an on-board algorithm validated against video-EEG. Empatica's own 2021 study calls it a predefined machine-learning algorithm. The McGrath event is known only from this single news item. The AES case-study abstract is not among the sources.

Comparison

seizures recorded during more than 6,500 hours of monitoring of 135 patients (device test reported by IEEE Spectrum)

After

the device correctly identified nearly every single seizure, with a false alarm rate below that of the seizure frequency for most individuals

Comparison basis

Measured

Full case and sources

AI system

AI-based Intrusion Detection System (distributed acoustic sensing on optical fiber), Northeast Frontier Railway

Deployer

Northeast Frontier Railway (Indian Railways)

AI contribution

The operator's release says the drivers noticed the herd and applied the emergency brake and were “also alerted” by the IDS. One local report says the braking followed the IDS warning. The AI alert is documented, but its share of the causal credit is not separable from the crew's own sighting. Hence protective response, prevention unproven.

Comparison

train-elephant collisions before the system (national average of elephant deaths on tracks)

Before

on an average 20 elephants die due to train collisions in the country every year and a majority of these incidents take place in the Northeast Frontier Railway

After

since the launch of the system, these 11 corridors have not reported any train-elephant collision

Comparison basis

Measured

Full case and sources

AI system

ChatGPT with OpenAI's safety-monitoring pipeline (automated classifiers plus human review and law-enforcement referral)

Developer

OpenAI

Deployer

OpenAI

AI contribution

OpenAI's automated classifiers and human reviewers flagged the user's messages, and the company chose to refer the account to the FBI. The system did not predict a crime or act on its own. The warrants, the arrest and the evidence-gathering were police and court actions. OpenAI sent only the user's prompts, not ChatGPT's replies. What the assistant said to the user, and whether it helped or hindered him, is unknown.

Full case and sources

AI system

Saronic Corsair unmanned (autonomous) surface vessel operated by U.S. 5th Fleet's Task Force 59, Corsair; navigation-software version not reported

Developer

Saronic Technologies (Texas)

Deployer

U.S. Navy, 5th Fleet Task Force 59 (U.S. Naval Forces Central Command)

AI contribution

Hawkins identified an unmanned Corsair as the recovery vessel. DefenseScoop describes the platform’s autonomous capabilities, but its suggestion that passive sensing helped locate the crew is an inference. The official sources do not identify the navigation mode, human supervision, or sensor contribution during this rescue. A crewed helicopter completed the recovery.

Full case and sources

AI system

Asilla AI behavior-analysis system for security cameras

Developer

Asilla Inc.

Deployer

About 40 railway stations and commercial buildings in Japan (operators not named)

AI contribution

Kyodo attributes both interventions to Asilla’s account. The company’s original product documentation supports the existence and intended mechanism of jump-risk alerts, but does not independently verify either intervention. The facilities, dates, alert logs, and subsequent outcomes are not supplied.

Full case and sources