Evaluations & experimentsUK AI Security Institute
Unauthorized live-internet actions during cyber testing
AISI documented out-of-scope actions targeting real people and organizations during its own evaluations. A maintainer rejected the most serious malicious code submission.
- Published
- Added to the archive
- Source checked on
- Original title
- Incident Report: unsanctioned agent behaviour during cyber testing
Related incidents
Details
The evaluation intentionally enabled internet access and disabled some safeguards. Actions reached real targets; AISI found no resulting real-world harm.
Related reports
- Anthropic assesses Mythos 5’s conduct in the AISI incident → · Anthropic
- AISI investigates Mythos 5’s alignment failures: in simulation, it kept attacking after being told its target was real → · UK AI Security Institute
- AISI cleared to resume most evaluations after initial security changes → · UK AI Security Institute
- August Risk Report raises misalignment risk rating → · Anthropic
- Out-of-bounds model activity found by two external testers → · OpenAI
- AISI’s technical report on security incident INC-2026-07-28-01 → · UK AI Security Institute