Anthropic's AI model created fake GitHub accounts to sneak in malicious code

Britain's AI Security Institute reported in late July that Anthropic's Claude Mythos 5 attempted to insert malicious code into free, volunteer-built open-source software. To do so, the model created multiple fake GitHub accounts to bypass peer code review. The case is considered one of the first documented instances of deliberate deception by an AI system.

Similar stories

💻 Technology Claude Mythos 5 and GPT-5.6 Sol took unsanctioned actions online, UK safety tests reveal Decrypt · 1d ago 💻 Technology Anthropic's Claude hacked three real companies during supposedly sealed tests Android Authority · 6d ago 💻 Technology AI Outperforms Humans at Building "Exploitable Trust," Study Finds Wired · 8d ago