Anthropic has integrated its advanced Claude Mythos 5 model into Claude Security, enabling enterprise teams to perform vulnerability scanning without directly accessing the model. As of August 21, 2026, Claude Security scans utilize the Mythos-class model, previously available only to vetted defenders through Project Glasswing. This scanning process connects to a GitHub repository, traces data flows across files, and provides users with findings categorized by Common Weakness Enumeration (CWE), along with confidence and severity ratings and suggested patches.
Claude Security is currently in public beta for Claude Enterprise customers, allowing users to initiate scans at claude.ai/security without requiring a separate model add-on. The scanning service is billed as standard token usage under the existing enterprise plan. Only organizations with a Claude Enterprise agreement and code stored in GitHub can access these scans; startups and mid-market teams on lower-tier plans are not eligible.
Industries that can benefit from this service include healthcare, financial systems, utilities, and the software supply chain. Anthropic has identified these sectors as key users for its security scanning capabilities. The application of the scanning technology spans pre-release code audits, legacy code reviews, scheduled regression scanning, and managing existing vulnerability backlogs.
Stay Ahead of the Curve!
Don’t miss out on the latest insights, trends, and analysis in the world of data, technology, and startups. Subscribe to our newsletter and get exclusive content delivered straight to your inbox.
In addition to the scanning capabilities, Anthropic announced the launch of the Defender Advantage Fund, providing $35 million in Claude credits aimed at organizations focused on securing open-source software. Furthermore, the company plans to expand its Cyber Verification Program, which will enhance dual-use capabilities on its other models, Opus and Sonnet, with access to Mythos-class features expected to follow.
The scanning mechanism operates by connecting to a GitHub repository, where Claude Mythos 5 analyzes the code, tracing data flows and reviewing Git history instead of relying solely on pattern matching. Each finding undergoes an adversarial verification step, in which the model assesses its own results to minimize false positives. The output includes a CWE category, confidence and severity ratings, and suggested fixes, which can be communicated via Slack or Jira or exported in CSV or Markdown formats.
Patching is conducted separately through Claude Code, where users implement fixes using models already available in their accounts. The integration is designed to ensure that while Mythos identifies vulnerabilities, it does not facilitate exploit creation since users receive fixed outputs rather than interactive prompts. This approach aims to mitigate risk by preventing users from steering the model towards exploit generation.
Key points from this update include the deployment of Claude Security scans running on Mythos 5, which are now available to Claude Enterprise customers in public beta. There is no need for separate model access; scans are part of existing plans. Findings are delivered with detailed categorizations and recommendations for human approval, ensuring thorough oversight in the patching process.