Anthropic's open-source framework for AI-powered vulnerability discovery

Anthropic has released an open-source reference implementation called Defending Code Reference Harness that enables autonomous vulnerability discovery and remediation in source code using Claude AI. The framework follows a recon → find → triage → report → patch workflow and can be customized for different programming languages and vulnerability types, while running in a secure gVisor sandbox to prevent malicious code execution. Anthropic also offers Claude Security, a managed product version that automates vulnerability scanning, verification, and fix generation across multiple projects.