Anthropic reconfigures its cool kids security program

After warning about the capabilities of competitor Z.ai's GLM-5.3 model, Anthropic has expanded its Cyber Verification Program (CVP) – now reconfigured into a three-tier system. The company previously offered its services through two programs: Project Glasswing and the CVP, which launched alongside its Mythos model in April 2026.

Now, these programs have been merged, with the aim of granting more security organizations access to Anthropic's security features. In total, Anthropic's security program has identified over 129,000 verified software vulnerabilities between April and July 2026. Of these, more than 33,000 have been rated as critical or high severity.

However, only 516 of the 5,674 true positive vulnerabilities have been patched. Anthropic's new CVP tiers include Defense Access for system defense, Red Team Access for penetration testing and offensive cyber evaluation, and Specialized Access reserved for verified organizations testing safety systems impacting people's lives or disrupting markets.