Anthropic has activated its security scanner, Claude Security, on the Claude Mythos 5 model for Enterprise customers in public beta. The tool scans codebases for vulnerabilities and suggests patches while counting as standard token usage. Each finding includes a CWE category, severity ratings, and a suggested fix, though humans must sign off on every patch before deployment. This rollout aims to boost defenders without providing attackers with new AI-powered options. The model is not broadly available because it is specifically tuned for cyber tasks where precision outweighs general capability. Using the most powerful model for this specific job ensures the scanner catches complex flaws that simpler tools might miss. Security teams can now integrate these detailed reports directly into their existing workflows without changing their token budget structures.
The practical impact lies in reducing the manual effort required to audit large codebases for known weaknesses. By automating the identification and patching suggestions, organisations can address vulnerabilities faster while maintaining strict human oversight. This approach balances efficiency with safety, ensuring that powerful AI does not inadvertently create new attack vectors.
- Scans run on the most capable Anthropic model currently available
- Human approval is mandatory for every suggested patch
- Results include standard CWE classifications and severity levels
