Anthropic opens Claude Security public beta for code audits

Anthropic has launched Claude Security into public beta, a dedicated defensive tool powered by its flagship Opus 4.7 model that scans entire codebases for vulnerabilities, generates verified patches, and traces data flows across components rather than relying solely on known attack pattern matching, SiliconANGLE reports.

The capability, now part of the Claude Enterprise subscription tier following a research preview that began in February, reasons "similarly to the way a cybersecurity researcher does," according to the company, examining interactions between code files and synthesizing network effects before assigning confidence ratings to each finding. Anthropic has attracted integration commitments from major cybersecurity vendors who will embed Opus 4.7 into their platforms. The tool provides explainability alongside its triage, documenting reasoning, exploitability likelihood, and fix effectiveness, and allows users to apply patches directly within a Claude Code session without the traditional days-long back-and-forth between security and engineering teams.

Building on the momentum of the earlier Project Glasswing initiative, which leveraged the company's Mythos model for vulnerability discovery across critical software, the public beta adds scheduled scanning for continuous coverage, documented dismissal reasoning for future auditors, and export compatibility with existing audit infrastructure.