Claude Mythos 5 Goes to Cyberdefenders via Project Glasswing, ~150 Organizations in 15+ Countries
Anthropic is deploying Claude Mythos 5, billed as having the strongest cybersecurity capabilities of any model in the world, to cyberdefenders and infrastructure providers through Project Glasswing.
Anthropic is putting its most capable model directly into the hands of defenders. Alongside the public launch of Claude Fable 5 on June 9, the company announced that Claude Mythos 5 - which it describes as having the strongest cybersecurity capabilities of any model in the world - will be deployed to cyberdefenders and infrastructure providers through Project Glasswing, a program run in collaboration with the US government.
What Project Glasswing Covers
The Mythos 5 rollout upgrades organizations that were already using Claude Mythos Preview, and expands the program to roughly 150 new organizations across more than 15 countries. Access is restricted to cyberdefenders and infrastructure providers, with select biomedical researchers gaining access later through a separate trusted access program.
Claude Mythos 5 is the same underlying model as Claude Fable 5, but with safeguards lifted in some areas. That distinction is the core of Anthropic's two-track strategy: the public gets a Mythos-class model made safe for general use, while vetted defenders get fuller capabilities under controlled conditions.
Guardrails on the Public Side
For the publicly available Fable 5, Anthropic built cybersecurity classifiers that detect sensitive queries. When triggered, those queries are answered by Claude Opus 4.8 instead - a fallback that fires on less than 5% of sessions on average. The company also reported:
- More than 1,000 hours of external red-teaming, with no universal jailbreaks found.
- An external partner confirmed that zero harmful single-turn requests relating to planning a cyberattack succeeded.
- Biology and chemistry classifiers, plus distillation prevention that blocks capability extraction for training competing models.
Mythos-class models also carry a 30-day data retention requirement for business customer traffic, used for safety monitoring only - not training - with all human access logged.
A Calculated Bet
The timing drew attention: TechCrunch noted the release came days after Anthropic warned that AI is getting too dangerous. By steering its strongest offensive-adjacent capabilities toward defenders first, Anthropic is effectively arguing that the best response to increasingly dangerous AI is to arm the people protecting critical infrastructure - before adversaries catch up.
For the organizations inside Project Glasswing, the upgrade lands immediately. For everyone else, Claude Fable 5 - the same model with guardrails intact - is available starting today.