As of August 21, 2026, Anthropic has integrated its most cyber-capable model, Claude Mythos 5, into the public beta of Claude Security for Enterprise customers. This update allows organizations to scan GitHub repositories for high-severity vulnerabilities without granting direct interactive access to the model.
- Scans trace data flows across files and Git history rather than relying on pattern matching, with findings including CWE categories, confidence, severity ratings, and suggested patches.
- The interface returns fixed scan results instead of a prompt box, preventing users from steering the model to write exploits.
- Anthropic launched the Defender Advantage Fund (0xDAF), offering $35 million in Claude credits to secure open-source software.
- Mythos 5 runs only inside the scan; interactive patching uses the organization's existing account models and requires human approval.
This packaging ensures that while the model possesses frontier cybersecurity capabilities, the risk of misuse is mitigated by restricting interaction to predefined output artifacts rather than direct prompting.