Anthropic has launched a public beta of a security tool called Claude Security that detects vulnerabilities in enterprise codebases and generates patches, SiliconANGLE reported on April 30.

Claude Security is available to users of Claude Enterprise, a subscription product for businesses.

Anthropic said it began a research preview in February under the name Claude Code Security. It said hundreds of companies then used it to find vulnerabilities in production code and apply patches. It added that it also found vulnerabilities that existing tools had not detected for years.

Claude Security is based on Anthropic's flagship model Opus 4.6. Security companies including CrowdStrike, Palo Alto Networks, SentinelOne, Trend Micro's TrendAI and Wiz have integrated Opus 4.7 into their platforms.

Claude Security analyzes an entire codebase in a manner similar to cybersecurity researchers. Instead of looking for known patterns, it tracks data flows, reads source code and analyzes interactions among code components. It provides results with confidence scores. It also provides explanations of exploitability, classification criteria and the effect of fixes.

Users can open a Claude Code session and proceed directly to patching based on the analysis results. Anthropic said it can also cut review processes that used to take days between security and engineering teams.

Anthropic is also running Project Glasswing, which uses the Mythos model that it has made available only on a limited basis. Glasswing focuses on strengthening software security with Mythos in cooperation with multiple partners.