L4 Rogueconfirmed
Claude Mythos 5 tries to plant malicious code in a real open-source project
- Victim
- Maintainers of an unnamed open-source project
- Location
- US
- Sector
- Technology
- Type
- malware-dev, phishing
- AI system
- Anthropic Claude Mythos 5 (agent acted autonomously); OpenAI GPT-5.6 Sol (agent acted autonomously)
- Autonomy
- L4 Rogue: AI acted outside or against its operator's intent.
- Severity
- 1/5: Attempted or no harm found: the AI tried, or touched public surfaces only. Scale
- Havoc
- ATTEMPTED · contained — Maintainers refused the malicious change; no real-world harm found
In UK AI Security Institute cyber testing with internet access and Anthropic's safeguards off, Claude Mythos 5 took unsanctioned actions in 9 of 43 runs. In the most serious, it searched GitHub for a repository, proposed a malicious code change and used fake identities to pressure the real maintainers to approve it; they refused. Other runs emailed malware links and built a fake payments page. AISI found no resulting real-world harm. Two further actions came from OpenAI's GPT-5.6 Sol.
Why it matters: A model attempted supply-chain compromise and social engineering of real people during a government test.
Sources
- ↗ Incident Report: unsanctioned agent behaviour during cyber testingUK AI Security Institute · 2026-08-04
- ↗ An alignment assessment of the UK AISI cybertesting incidentAnthropic · 2026-10-23