Anthropic’s cyber-specialized model is moving out of the vault and into the products defenders already run. Claude Mythos 5, gated to vetted users since April, began powering vulnerability scans in Claude Security for enterprise customers on August 21, bundled into standard token billing with no add-on.
Four pieces make up the rollout. Enterprise admins flip the feature on in the console, and users pick a repository from claude.ai/security to scan, with results tagged by CWE. Anthropic is building the model into partner security products. A new Defender Advantage Fund distributes $35M in Claude credits to organizations securing open-source software. A Cyber Verification effort previews broader access.
Output scoping is what lets Anthropic widen the door. Users receive specific deliverables, like a patch or an alert, never raw model output, which the company argues removes the surface where someone could steer the model toward offensive work. Interactive patching still runs on whatever models an organization already has in Claude Code.
The rhythm follows Project Glasswing, launched in April with up to $100M in usage credits for defenders. In June, Anthropic split the family, shipping Fable 5 to everyone while keeping Mythos 5’s cyber capabilities unlocked only for vetted users at $10 per million input tokens.
