Expanding the Cyber Verification Program

Anthropic Formalizes Tiered AI Access for Cyber Professionals

Anthropic has officially expanded its Cyber Verification Program (CVP), creating a structured, multi-tier system to provide vetted security professionals with access to its most powerful models, including the new Claude Opus 5.5. The program is designed to address the inherent dual-use risk of advanced AI by equipping cyber defenders with state-of-the-art tools while maintaining robust safeguards on its generally available models. This move consolidates previous initiatives like Project Glasswing into a single, scalable framework intended to give more security organizations the capabilities needed to protect critical systems.

A Look at the CVP Tiers

The updated program introduces three distinct access tiers, each with specific verification requirements, security controls, and permitted use cases. This granular approach allows Anthropic to tune the model's blocking classifiers based on the user's validated role, moving beyond a one-size-fits-all safety policy. Performance evaluations on the CyScenarioBench benchmark showed that without CVP access, every task was blocked, while the Red Team Access tier experienced no blocks and achieved a 67.6% success rate, demonstrating the efficacy of the tiered safeguards.

• Defense Access: For defensive security operations, incident response, malware analysis, and vulnerability validation. The application review process is expected to take a few days.
• Red Team Access: Adds authorized penetration testing and red-teaming to the defensive uses. Applications for this tier are expected to take a few weeks to review.
• Specialized Access: Reserved for a limited set of verified organizations testing high-risk safety systems, such as power grids and flight operating systems, with the fewest cyber blocks.

The impact of providing controlled access has already been demonstrated through the program's precursor, Project Glasswing, where partners identified at least 129,000 verified software vulnerabilities between April and July 2026. By formalizing this access, Anthropic aims to systematically scale these defensive benefits across the industry. The CVP is available on the Claude Platform, Google Cloud Vertex AI, and Microsoft Foundry, with availability on Amazon Bedrock limited to customers eligible for the upcoming Enterprise Frontier Safeguards (EFS) solution, which will offer zero data retention.

Anthropic's tiered CVP framework establishes a critical industry precedent for managing the dual-use nature of frontier AI models, shifting the focus from universal restrictions to verified, role-based access for specialized professional domains.