Claude Security Now Scans Continuously and Tracks Fixes Until Closed
Claude Security now scans repositories continuously and tracks each finding until a fix ships. Anthropic announced the update on October 21, 2026, in public beta for Claude Enterprise customers. Anthropic says that since the product entered public beta in April, thousands of organizations have used it to find more than a million vulnerabilities.
The mechanics matter more than the headline. Teams create a project, connect repositories through the Claude GitHub app (including GitHub Enterprise Server) or a GitLab or other Git token, and choose branch and scope. Claude runs a baseline scan, then rescans changes by risk and complexity. A project can span multiple repositories, which targets issues that arise between services rather than inside one. Fix tracking checks open findings against changed code, marks them fixed when a fix ships and reopens them if the issue returns. Each project has a knowledge base for deployment setup and security policies, and Claude generates an editable threat model that guides scans and severity ratings. An API, webhooks, an MCP connector and role-based access controls cover integration. Organizations receive $25,000 in monthly free usage credits for two months, applying to usage through December 31, 2026.
The harder question is what happens after the scan. Comcast used Claude Mythos to assess 258 business-critical systems containing roughly 170 million lines of code and found an authentication bypass that arose from interactions across multiple systems. CISO Noopur Davis said “validation of these volumes of findings is the new bottleneck”. Fix tracking addresses the same problem: it turns a pile of findings into a queue with state.
For leaders weighing AI-assisted security, use the credit window to test three things: how many findings your team can validate per week, whether the generated threat model reflects your actual policies, and whether reopened issues surface in the tools your engineers already use. Measure validated, closed findings, not findings raised.