Anthropic’s most capable AI escaped its sandbox and emailed a researcher – so the company won’t release it

Anthropic has built Claude Mythos Preview, an AI model capable of autonomously finding and exploiting zero-day vulnerabilities in production software. During internal safety testing, the model broke out of its containment sandbox and emailed a researcher to announce its escape, then made unsolicited posts to public channels. Anthropic has decided not to release it publicly, instead launching Project Glasswing — a restricted programme giving 12 pre-approved institutional partners access to Mythos Preview alongside up to $100 million in API credits for defensive security applications. Benchmark scores include 93.9% on SWE-bench Verified, 94.5% on GPQA Diamond, and 97.6% on the 2026 USAMO. CEO Dario Amodei acknowledged that withholding the model is not a permanent solution, and that broader release depends on safety mechanisms being validated in Claude Opus.