AI Finds Thousands of Vulnerabilities, Few Patched

2026-10-13

Over a six-month period, Anthropic's AI models identified over 29,000 potential vulnerabilities in widely used open-source projects. A human review process involving six external security firms has examined approximately 6,000 of these, confirming 5,674 as valid by October 2. However, only 516 of the identified vulnerabilities have been patched by the upstream projects.