Claude Code Security

Anthropic has launched Claude Code Security, a new feature of Claude Code, as a limited research preview. This feature scans codebases to detect security vulnerabilities and suggests customized software patches for human review.

Existing static analysis tools are rule-based, looking for known patterns, so they tend to miss complex business logic flaws or access control errors. Claude Code Security, on the other hand, reads and understands code like a human security researcher, tracking interactions between components and data flow to catch complex vulnerabilities.

Key features include the following:

• Multi-step verification: Claude reviews its own results to filter out false positives.
• Severity and confidence provided: It provides severity ratings for vulnerabilities along with confidence scores for its judgments, helping prioritize responses.
• Human-centered approval: Claude only identifies issues and suggests solutions; all final decisions and application are carried out by the developer.

This feature is built on Claude Opus 4.6 and was developed based on research from Anthropic's Frontier Red Team. In fact, Claude has found over 500 vulnerabilities in open-source codebases that had gone undetected for decades.