AI Spots Code Flaws Before Hackers Can

Anthropic has introduced Claude Security, a new AI-powered cybersecurity product now available in public beta for Enterprise-tier Claude users, with broader availability soon. This tool leverages the Claude Opus 4.7 model to scan codebases for vulnerabilities and generate targeted patches, aiming to proactively defend against cyberattacks.

Claude Security functions by analyzing code to identify potential weaknesses, much like security researchers. It traces data flows, reads source code, and understands component interactions across files and modules. The system includes a multi-stage validation pipeline that independently verifies each finding, assigning a confidence rating and providing detailed explanations, including severity and recommended fixes.

The AI's findings can be immediately acted upon, allowing developers to open the relevant code in Claude Code for in-context modifications. Workflow optimizations such as scheduled scans and the ability to dismiss findings with documented reasons are also included. This allows security teams to focus on high-confidence, large-impact issues first.

Anthropic is also implementing cyber safeguards within its Opus 4.7 model to prevent misuse. These safeguards automatically detect and block requests associated with prohibited or high-risk cybersecurity activities. Approved researchers in Anthropic's Cyber Verification Program can, however, access restricted capabilities for legitimate defensive research.