Anthropic Launches Claude Security Public Beta to Help Enterprises Detect AI Code Vulnerabilities

Anthropic has opened the public beta of Claude Security to all Claude Enterprise users. Built on the Claude Opus 4.7 model, enterprises don't need to build custom agents or API integrations—just point to a GitHub repository to start scanning.

Unlike rule-based static analysis tools, Claude Security reasons about code like a human researcher, understanding component interactions and data flows to identify complex vulnerabilities. Each finding goes through multi-stage validation; the model self-questions to reduce false positives and rates severity based on actual exploitability.

The previous Opus4.6 model found over 500 hidden vulnerabilities in production open-source code repositories that experts missed for decades. During closed testing, hundreds of organizations used the tool to find vulnerabilities missed by existing tools. The beta supports periodic scans, specific directory limits in large monorepos, and results can be exported as CSV/Markdown or pushed to Slack/Jira via Webhooks.