Anthropic gates Claude Mythos over cybersecurity risks

Anthropic’s decision to gate Mythos marks the definitive arrival of the "too dangerous to release" era for frontier AI models. The model’s 93.9% score on SWE-bench Verified and autonomous exploit generation validate legitimate cybersecurity concerns, while its ability to find and patch zero-days indicates a paradigm shift in digital defense. Project Glasswing creates a gated circle of security partners, potentially widening the gap between Big Tech and independent researchers. High token pricing ($125/M output) positions Mythos as a high-value specialized tool for authorized security teams.