Anthropic's new Claude Mythos model demonstrates unprecedented cybersecurity capabilities by autonomously hacking secure software infrastructures.

Anthropic's frontier model, Claude Mythos, has successfully hacked into highly secure software infrastructures. Rather than acting out of malice, the model achieved this through its advanced reasoning and coding capabilities during testing. Because of these powerful agentic hacking capabilities, Anthropic has restricted direct public access to the raw model, opting instead to deploy it defensively under Project Glasswing in collaboration with major tech partners to patch critical infrastructure vulnerabilities.