Anthropic is expanding access to its Cyber Verification Program (CVP), offering vetted security professionals and organizations access to advanced Claude AI models with reduced safety restrictions. The initiative allows qualified researchers to perform vulnerability research, malware analysis, incident response, and penetration testing, which are typically blocked by standard public guardrails.
The CVP structure introduces three distinct access levels: Defense Access for broad security teams, Red Team Access for authorized organizational attack simulations, and Specialized Access for critical infrastructure like power grids and flight controls. Specialized Access applications are evaluated in coordination with the U.S. government to manage risk.
According to Anthropic, partners participating in its predecessor program, Project Glasswing, identified at least 129,000 confirmed vulnerabilities between April and July 2026, with over 33,000 rated as high-severity or critical. Partners reported that AI capabilities accelerated vulnerability identification workflows by months or years.
Why it matters
Provides security teams dedicated AI tooling with reduced guardrails for legitimate defensive testing.
Demonstrates formal alignment between frontier model labs and government oversight for critical infrastructure safety.
Quantifies AI efficiency gains in identifying software vulnerabilities and managing patch lifecycles.
Source: the-decoder.com



