Anthropic introduces claude security with mythos 5 for enterprise vulnerability detection

Anthropic has launched a new security feature called Claude Security, which leverages its advanced Mythos 5 model to detect vulnerabilities in enterprise systems. The tool is now available in public beta for Claude Enterprise customers, allowing security teams to scan GitHub repositories for potential security issues without direct access to the model. The feature was announced on August 21, 2026, and operates by tracing data flows across files and analyzing Git history to identify risks. Each finding undergoes an adversarial verification process to reduce false positives, and results include CWE categories, confidence levels, severity ratings, and suggested patches. Results can be shared via Slack or Jira, or exported as CSV or Markdown. Anthropic emphasized that the model does not provide direct access to the underlying Mythos 5 capabilities, ensuring that vulnerabilities are identified without the risk of exploit generation. The company is also collaborating with cybersecurity vendors to integrate Mythos 5 into third-party tools. Patching is handled separately through Claude Code, which uses existing models within the organization. All patches require human review and approval. Anthropic highlighted that Mythos 5 represents the strongest cybersecurity capabilities among its models, with the ability to both find and potentially write exploits. The release focuses on controlling access to the model’s capabilities through a structured interface, ensuring that users receive predefined outputs rather than direct model interaction. The feature is part of a broader strategy to enhance enterprise security without compromising model safety.