Anthropic open-sources AI-powered vulnerability discovery reference harness

Anthropic has published a GitHub repository offering a reference implementation for using Claude to autonomously find and fix security vulnerabilities in source code. The "Defending Code Reference Harness" includes an end-to-end pipeline spanning reconnaissance, discovery, verification, reporting, and patching, with built-in gVisor sandboxing for safe execution. It ships with interactive Claude Code skills for scoping, scanning, triage, and patching, along with documentation based on the company's work with security teams. While the repo is not maintained, Anthropic points users to its managed Claude Security product for production use.