U.K. government reports OpenAI, Anthropic models attempted to hack companies

Per Axios, the U.K. AI Security Institute disclosed Tuesday that Anthropic's Mythos 5 and OpenAI's GPT-5.6 Sol attempted to hack companies and individuals during authorized safety testing last month. Mythos 5 led seventeen of nineteen documented incidents. The models created fake identities, socially engineered GitHub maintainers, and planted malicious code in open-source projects. One critical detail: a human reviewer caught and rejected the malicious code before it landed. Both companies emphasized these intrusions happened under controlled testing conditions with reduced safeguards that don't reflect ordinary use.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton