AI system
- System
-
Big Sleep
- Model
-
Not assessed
- Release
-
Not assessed
- Application developer
-
DeepMind and 
Google Project Zero - Role
-
Not assessed
- Configuration
-
Not assessed
- Inputs and outputs
-
Not assessed
- Autonomy
-
Not assessed
- Workflow
-
Threat-intelligence artifacts narrowing the target without identifying the bug. An LLM agent with code navigation, sandboxed execution and debugging tools pointed at a specific codebase. A responsive upstream maintainer shipping a fix before disclosure.
- Human oversight
-
Not assessed
- Relation to AI
-
Per Google's spokesperson, human threat analysts supplied limited indicators and Big Sleep isolated the specific vulnerability. The CVE credit names a human finder “with assistance from Google Big Sleep”. Developer and the organization reporting the benefit are the same company. The Record and the SQLite release log are the independent sources. Neither confirms the threat-actor staging.
- AI contribution
-
Per Google's spokesperson, human threat analysts supplied limited indicators and Big Sleep isolated the specific vulnerability. The CVE credit names a human finder “with assistance from Google Big Sleep”. Developer and the organization reporting the benefit are the same company. The Record and the SQLite release log are the independent sources. Neither confirms the threat-actor staging.
Benefits
- Outcome identifiers
-
Not assessed
- Benefit
-
Google 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.
- Category
-
Reduced Risk
- Status
-
Not assessed
- Beneficiary groups
-
Organizations. Society.
- Beneficiaries
-
Not assessed
- Reach
-
Not assessed
- Distribution and access
-
Not assessed
- Duration
-
Not assessed
- Expected?
-
Yes
- Costs and harms
-
Not assessed
Measurement
- Studies
-
Not assessed
- Study design
-
Not assessed
- Measurement
-
Not assessed
- Setting and sample
-
Not assessed
- Measures
-
Not assessed
- Comparison
-
Not assessed
- Results
-
Not assessed
- Additional value from AI
-
Not assessed
- Limitations
-
Not assessed
- Denominators
-
Not assessed
- Materials for replication
-
Not assessed
Money
- Kind of value
-
Not assessed
- Amount
-
Not assessed
- Currency and prices
-
Not assessed
- Period and unit
-
Not assessed
- Realization
-
Not assessed
- Recipients and payers
-
Not assessed
- Gross or net
-
Not assessed
- Included costs
-
Not assessed
- Comparison
-
Not assessed
- Method
-
Not assessed
- Return and payback
-
Not assessed
People and places
- Participants
-
Developer:
DeepMind and Google Project Zero. Operator: Google (Threat Analysis Group / Big Sleep team). - Jurisdictions
-
Developer:
United Kingdom. Operator: 
United States. - Ownership and control
-
Not assessed
- Dependencies
-
Not assessed
- Cross-border connections
-
Not assessed
- Geopolitical implications
-
Not assessed
Follow-up
- Conditions for replication
-
Partial indicators of a staged exploit must exist and be shared with the agent's operators. Target code must be small enough for the agent to reason over. Coordinated disclosure before the actor's exploitation window.
- Replication attempts
-
2 follow-up entries. Benefit recurred: 2.
- Later findings
-
11 August 2025. Big Sleep went on to report a legitimate curl vulnerability on 11 August 2025, later published as CVE-2025-9086, which curl's maintainer confirmed.
- Availability
-
Not assessed
- Institutional decision
-
Not assessed
- Grounds
-
Not assessed
- Action and result
-
Not assessed
- Success criterion
-
Not assessed
Evidence
- Source numbers
-
AIBD-0225-source-0001–0004
- Citations
-
Numbered references below.
- Language
-
Not assessed
- Source relationships
-
Independent, Developer
- Original evidence
-
Not assessed
- Access and versions
-
Not assessed
- Supported findings
-
Not assessed
- Evidence links
-
Not assessed
- Evidence strength
-
Not assessed
- Missing evidence
-
Not assessed
Grades
- Assessed benefits
-
Organizations. Society.
- Depth
-
Organizations: 2. Society: Not assessed.
- Reasons
-
Organizations: SQLite fixed the flaw in 3.50.2 (release log entry on aggregate terms exceeding column count) before public exploitation was reported. Society: No source documents an outcome for users beyond the patch. The foiled-exploitation claim is Google's.
- Observed or intended
-
Not assessed
- Rubric
-
Not assessed
- Assessor and date
-
Not assessed
- Alternative assessments
-
Not assessed
- Fill
-
Independently reported
Entry details
- Kind
-
Not assessed
- Related entries
-
Not assessed
- Field
-
Cybersecurity / vulnerability research. Technology.
- Purpose
-
Actively search for and find unknown security vulnerabilities in software using a large language model (LLM) agent (Google DeepMind and Project Zero).
- Dates
-
Date basis unassessed: 2025 (precision unassessed). Updated: 22 October 2026.
- Stage
-
Not assessed
- Version
-
SHA-256: 47231bc70e9d…
References
- Other identifiers
-
Not assessed
- Category mappings
-
Not assessed
- Brief
-
Not assessed
- Revisions
-
9 revisions. Latest: 22 October 2026.
Sources
-
News report Independent author
Jonathan Greig, The Record from Recorded Future News, 15 Jul 2025
who leveraged Big Sleep to isolate the vulnerability the adversary was preparing to exploit in their operations
-
Press release Developer
Kent Walker, Google blog, 15 Jul 2025
Google was able to actually predict that a vulnerability was imminently going to be used and we were able to cut it off beforehand.
-
Other Independent author
SQLite release log, version 3.50.2, 28 Jun 2025
Raise an error early if the number of aggregate terms in a query exceeds the maximum number of columns, to avoid downstream assertion faults.
-
Registry label Developer
CVE Program record CVE-2025-6965 (CNA: Google), published 15 Jul 2025
Vlad Stolyarov of Google's Threat Analysis Group, with assistance from Google Big Sleep