AI Reports · Anthropic · 2026-10-20 · Blog / commentary
Using Claude to reduce the attack surface of code
Anthropic previews its initial research on automated attack surface analysis and reduction, which uses AI models to map reachable code and then remove or restrict access to anything a given deployment doesn't need to expose. Reducing code reachability can lower the severity of unknown vulnerabilities without having to find or fix each one individually. In a preview of data from large open-source codebases, Claude ran over 25 projects in a variety of languages and discovered 493 issues.