What Is Claude Mythos and Project Glasswing

Claude Mythos marks Anthropic’s most capable AI tier, positioned above Claude Opus, and emerged amid a March 2026 data leak that brought Mythos to light. The company released Claude Fable 5, a safeguarded version, on June 9, 2026, for a brief public window, before a June 12 export-control order by the U.S. government forced its removal and the worldwide suspension of both Fable 5 and Mythos 5. The Mythos family is connected to Project Glasswing, with Mythos Preview limited to vetted cybersecurity partners and not generally available. In benchmarks, the original Mythos Preview scored 93.9% on SWE-bench Verified and helped identify more than 10,000 high- and critical-severity software vulnerabilities. Anthropic also notes Mythos 1 is being rolled into Claude Code and Claude Security for enterprise customers, signaling a broader shift to enterprise deployments while managing safety concerns.