Anthropic launches Mythos 5 in Claude Security for enterprises

Effective 21 August 2026, Claude Enterprise customers can run security scans powered by Mythos 5, the strongest cyber-capable model from Anthropic, which until now was only available to protectors evaluated under Project Glasswing. The scan connects to a GitHub repository, tracks data flows between files, and returns findings with a CWE category, security and severity ratings, and a suggested fix. The model is returned as a scan result rather than an open prompt, so the same capabilities that discover exploitable weaknesses cannot be directed to write exploits. The feature is in public beta without a separate model addition; using Mythos 5 is charged as regular tokens under the existing plan.

A user links a GitHub repository, and Mythos 5 scans it by tracking data flows between files and reading Git history rather than matching rule patterns. According to Anthropic, “each finding undergoes an adversarial verification step in which the model challenges its own result before it is presented, a mechanism intended to reduce false positives.” Findings are returned with a CWE category, security and severity ratings, and a suggested fix; the stated target is high-severity issues such as memory corruption, injection flaws, authentication bypasses, and cross-file logic errors. Results can be pushed to Slack or Jira via webhooks, or exported as CSV or Markdown. Scans can be scoped to a directory and scheduled.

The announcement bundled three items: first, Claude Security scans now run on Mythos 5 for Enterprise customers; second, Anthropic launched the Defender Advantage Fund (0xDAF), which offers $35 million in Claude credits to organizations that secure open-source software; third, the company said it will broaden its cyber-verification program in the coming weeks to cover broader dual-use capabilities in Opus and Sonnet, with a Mythos-focused department to follow. Anthropic is also working with cyber-security vendors to embed Mythos 5 in partner tools, and has opened an expression of interest form.

Mythos-department models sit above the Opus tier and have been withheld from release for that reason. Anthropic stated that Mythos 5 has the strongest cyber capabilities of any model in the world, and that those same skills that find vulnerabilities could also be used to write exploits. Claude Fable 5 is the same base model that is widely deployed, with classifiers that route queries in cyber, biology, chemistry and beyond. The argument in this launch is that the packaging—a scanning product that returns a result rather than a prompt box—enables the capability to be released without releasing the risk. Fixes are applied separately: users open Claude Code on the web to implement patches, and this step uses models the organization already has; a Mythos scan does not expand Mythos access to other surfaces. Every fix requires human review and approval.