Anthropic Releases Open-Source Framework for AI Vulnerability Discovery

Anthropic open-sourced a framework that leverages large language models to identify and validate security vulnerabilities in codebases. The tool, called Defending Code Reference Harness, uses adversarial techniques to probe for weaknesses, reducing manual effort. Initial benchmarks show it can find SQL injection and XSS flaws with high recall. This marks a shift toward AI-driven proactive security, though experts caution against over-reliance on automated findings.