LLMs Drive 3.5x Surge in Critical CVE Discoveries
Epoch AI reports that using LLMs for automated bug hunting has led to a 3.5x increase in critical and high-severity vulnerability discoveries. In a single month, 21 organizations reported 1,500 vulnerabilities, with analysts linking the surge to the release of Mythos—Anthropic’s security-focused initiative. Partners in the Glasswing program have collectively identified over 10,000 vulnerabilities, some still undisclosed, while Epoch AI estimates that OpenAI’s Daybreak initiative contributes a comparable volume.