Autonomous systems found real zero-days in DARPA AI Cyber Challenge
The autonomous systems found real, previously unknown zero-day vulnerabilities beyond the planted set the competition intended.
- What happened
- At the DARPA AI Cyber Challenge finals, announced on 8 August 2025, autonomous AI-driven cyber-reasoning systems found 54 of 63 planted (synthetic) vulnerabilities and patched 43, and additionally discovered 18 real, previously unknown zero-day vulnerabilities across C and Java open-source code, providing 11 patches, at an average cost of about $152 per task. Team Atlanta won $4 million, Trail of Bits $3 million and Theori $1.5 million. Finalist teams open-sourced their systems.
- Response
- Systems were released as open source under an OSI-approved licence, along with the competition framework, challenges and tooling, to support defensive use.
- Discovered by
- Competition results reported by DARPA at DEF CON.
- Who was affected
- Open-source software containing planted and previously unknown vulnerabilities (defensive research)
- Patterns
- Sources
- AI Cyber Challenge (AIxCC) ResultsDARPA, 8 Aug 2025
Related
Incidents with the same pattern
What does this ask of people?
Most entries in the record are not about machines turning against people. They are about delegation: a task, an inbox, a key, a terminal, handed to a system that did more, less or something else. Futures literacy starts with noticing that.