An opt-in vulnerability-finding service for open-source software

Anthropic is launching OSS Scanner, an opt-in, LLM-powered vulnerability scanner for open-source projects that produces fully model-generated security reports. The blog describes that their models discovered over 29,000 candidate vulnerabilities (roughly 6,000 manually triaged), produced some validated high-severity findings including chained unauthenticated RCEs, and offers projects the option to receive raw automated reports or follow coordinated vulnerability disclosure; enrollment is via a GitHub PR and eligibility mirrors OSS-Fuzz criteria.