Anthropic Expands AI Model Access for Cybersecurity Teams

Anthropic is broadening a program that gives vetted cybersecurity professionals access to its AI models with fewer safeguards. The initiative, called Project Glasswing, identified over 129,000 verified software vulnerabilities between April and July 2026, with an additional 5,500 found through open-source scanning. More than 33,000 of these vulnerabilities were rated as critical or high-severity, though the total impact is likely much higher. The Cyber Verification Program (CVP) offers three access tiers for organizations to conduct defensive work, penetration testing, and specialized safety system testing. Evaluations showed that the AI models’ safeguards effectively blocked tasks in the Defense Access tier but not in the Red Team Access tier, indicating a balance between security and functionality.