Anthropic Brings Claude Mythos 5 to Claude Security: Enterprise Teams Get Frontier Vulnerability Scanning Without Direct Model Access

Anthropic has moved Claude Mythos 5 into Claude Security, but only as a background scanner for Claude Enterprise customers. The feature went into public beta on August 21, 2026, and it is aimed at teams with code stored in GitHub. Startups and mid-market customers on Pro, Max, or Team plans don’t get this version of the scan.

The pitch is simple: point the product at a repository and it traces data flows across files, reads Git history, and returns findings with a CWE category, confidence, severity, and a suggested patch. Anthropic says the model also runs an adversarial verification step before it surfaces a result, with the goal of cutting false positives. It’s meant for the ugly stuff — memory corruption, injection flaws, authentication bypasses, and cross-file logic errors.

The important part is what Anthropic does not expose. Users get a scan result, not a prompt box, and there is no way to ask Mythos 5 for an exploit. Patching happens separately in Claude Code on the web, using whatever models the organization already has. Every fix still needs human review and approval. So the model that finds the bug is not the model you talk to about abusing it.

Anthropic also bundled in a few other moves: a $35 million Defender Advantage Fund in Claude credits for open-source security work, and an expansion of its Cyber Verification Program over the coming weeks to cover broader dual-use capabilities on Opus and Sonnet, with Mythos-class access to follow. There’s also an interest form for vendors that want to embed Mythos 5 inside their own security tools. The packaging matters here almost more than the model itself. Anthropic is betting that locked-down output is safer than open-ended prompting, which is a very neat theory right up until someone forgets the lock.