Anthropic Cyber Verification Program: What Changes for Security Teams

Anthropic has restructured how cybersecurity professionals access its most capable artificial intelligence models. According to Olhar Digital, the company combined previous initiatives into a unified framework called the Cyber Verification Program, or CVP, designed to put powerful tools directly into the hands of defenders.

How does the new Cyber Verification Program work?

The updated CVP establishes three distinct access tiers, each requiring specific vetting processes conducted alongside the United States government. The Defense tier supports tasks like incident response and malware analysis for critical infrastructure operators and researchers. The Red Team tier enables authorized organizations to perform penetration testing. Finally, the Specialized tier imposes the fewest restrictions, granting access to a small group of vetted entities testing high-risk systems such as power grids and aviation networks.

What results did the previous security initiatives achieve?

The program expansion follows the success of Project Glasswing. Between April and July 2026, participating partners uncovered at least 129,000 verified software vulnerabilities. Out of those findings, over 33,000 were classified as high severity or critical. Anthropic estimates that the real-world impact could be up to five times greater, as those figures came from a survey covering only a fraction of the program's total partners.

What are the risks of deploying less restricted AI models?

Granting access to models with fewer safety guards creates an inherent dilemma. The same advanced capabilities that allow a system to pinpoint a software flaw and assist engineers in fixing it can potentially be weaponized to exploit that very same weakness. To mitigate this risk, Anthropic restricts its less-filtered models—such as the Claude Mythos and Opus families—strictly to organizations that pass rigorous background verifications rather than releasing them publicly.