Anthropic is reportedly releasing a public version of its Mythos model tomorrow, potentially called Claude Fable. Mythos has been restricted since April to 200 organizations through Project Glasswing, a cybersecurity consortium that includes NATO, Samsung, and the EU's cybersecurity agency. The model autonomously discovered over 10,000 critical zero-day vulnerabilities across every major operating system and browser in its first month. Anthropic delayed public release because Mythos can autonomously develop professional-grade cyberattacks. The public version will have "substantial guardrails" and won't be as permissive as what Glasswing partners access. It is reportedly dramatically better at long-horizon, multi-turn tasks.