Anthropic Expands Cyber Verification Program with New Models and Partnerships

On October 6, 2026, Anthropic launched an expanded version of its Cyber Verification Program (CVP), introducing three new cyber models: Claude Opus 5.5, Claude Sonnet 5.5, and Claude Mythos 5.1. This expansion is part of a collaborative effort with partners Booz Allen, Comcast, and Barclays. Between April and July 2026, partners uncovered 129,000 verified software vulnerabilities, with an additional 5,500 verified by Anthropic from April to October 2026. Of these, 33,000 vulnerabilities were rated as critical or high severity. During evaluations, 46 out of 50 tasks were blocked in the Defense Access tier, while 34 out of 50 tasks were successfully completed in the Red Team Access tier, resulting in a 67.6% success rate for Claude Opus 5.5 without safeguards. Anthropic expects many organizations conducting defensive cybersecurity work to qualify for the new tier, aiming to extend the impact of Project Glasswing to a broader set of cyber defenders. The initiative aligns with Anthropic's ongoing commitment to enhancing cybersecurity capabilities, following their previous investment in training 10,000 engineers by the end of 2027.