Anthropic launches open-source framework for AI vulnerability discovery

The framework, dubbed defending-code-reference-harness, is designed to help security researchers find vulnerabilities using AI. It is open-source and available on GitHub, providing a reference implementation for automated security testing.