Making frontier cybersecurity capabilities available to defenders

Anthropic has launched a limited research preview of Claude Code Security, a new feature built into Claude Code on the web that scans codebases for software vulnerabilities and suggests targeted patches for human review. Traditional static analysis tools typically rely on rule-based pattern matching, which often catches common issues but misses subtle, context-dependent flaws like business logic errors or broken access control. In contrast, Claude Code Security reads and reasons about code similarly to a human security researcher by tracing data movement and component interactions. Before any finding reaches an analyst, Claude re-examines the result through a multi-stage verification process to filter out false positives and assigns severity and confidence ratings. Developers retain final authority, reviewing dashboard findings and approving patches within their existing tools. Anthropic developed this capability following more than a year of cybersecurity research, including stress-testing by its Frontier Red Team and using Claude Opus 4.6 to find over 500 vulnerabilities in production open-source codebases. The limited research preview is currently open to Enterprise and Team customers, with free, expedited access available for maintainers of open-source repositories. Anthropic notes that while attackers will increasingly use AI to discover weaknesses, defenders moving quickly can utilize similar capabilities to patch code and raise the industry's security baseline.