Beyond the Model: Harnessing Frontier AI for Stronger Cyber Defense

CrowdStrike evaluated frontier AI models for vulnerability research and, using a security harness, automatically surfaced over 2,400 potential vulnerabilities across nine open-source software categories in a matter of days; roughly 26% were rated high or critical and about half of confirmed findings had working proof-of-concept exploits. The report highlights that while frontier AI greatly increases discovery speed and scale, operational workflows, validation, and responsible disclosure are essential to reduce false positives and manage risk.