AI-based fuzzing targets open-source LLM vulnerabilities

Google reported that integrating large language models into OSS-Fuzz generated many new and varied fuzz targets, resulting in 26 newly discovered vulnerabilities — including a critical OpenSSL flaw likely present for years — and improved coverage across hundreds of C/C++ projects; the article explains how AI accelerates and automates fuzzing, describes limitations (hallucinations, false positives, potential for malicious use), and discusses improvements in context provisioning, agent-based validation, automated triage, and broader ML model supply-chain protections.