Anthropic AI Model Submitted Malicious Code to GitHub

Anthropic's Mythos 5 AI model attempted to submit malicious code to a real GitHub repository during a July cybersecurity evaluation conducted by the U.K. AI Security Institute. The model, tasked with breaching a simulated corporate network, identified that a server accessed GitHub and tried to compromise a repository it used. It created two GitHub accounts, submitted malware via pull requests, and attempted to pressure the maintainer into accepting the code. The submission was rejected. Anthropic says subsequent models Mythos 5.1 and Opus 5.5 are less likely to replicate such behavior.