Anthropic Expands Cyber Verification Program with New Access Tiers

News

Anthropic has revamped its Cyber Verification Program, introducing three access tiers to offer tailored AI capabilities for cybersecurity professionals.

Anthropic has announced a significant expansion of its Cyber Verification Program (CVP), designed to provide tailored access to advanced AI models for cybersecurity professionals. The revamp integrates the earlier CVP and Project Glasswing initiatives into a unified offering with three distinct access tiers. These tiers—Defense Access, Red Team Access, and Specialized Access—allow organizations to choose capabilities best suited to their specific cybersecurity needs.

Evidence and context

The updated CVP, unveiled on October 6, 2026, expands access to Anthropic’s most advanced AI models, including Claude Opus 5.5, Claude Sonnet 5.5, and Claude Mythos 5.1. Security professionals can now apply for targeted tiers based on their operational scope, offering enhanced capabilities for tasks ranging from defensive malware analysis to authorized penetration testing.

According to Anthropic, the new structure addresses the dual-use nature of cybersecurity AI. While the company’s generally available models include safeguards to prevent misuse, such blocks can hinder legitimate defensive work. By creating tier-specific access, Anthropic aims to balance security with functionality. For example, the Defense Access tier supports defensive tasks like vulnerability validation, while the Red Team Access tier adds permissions for authorized adversarial testing. The highest tier, Specialized Access, is reserved for critical safety systems such as power grids or telecom networks, with applicants vetted in coordination with the U.S. government.

Testing has affirmed the efficacy of Anthropic’s safeguards. On the CyScenarioBench evaluation, which simulates complex cyber operations, models in the Defense Access tier experienced significant blocking, while those in higher tiers—Red Team and Specialized Access—performed tasks with no interruptions, demonstrating the program’s ability to tailor capabilities securely.

Previously, CVP and Project Glasswing operated separately, with the former covering reduced safeguards for Claude Opus and Sonnet models, and the latter focusing on critical software security using Claude Mythos. The integration into a single, tiered model simplifies access while broadening eligibility. Additionally, Anthropic plans to introduce Enterprise Frontier Safeguards later this year, enabling cloud-controlled data retention for eligible organizations.

The program has already shown measurable impact. Between April and October 2026, Anthropic reported its models helped uncover over 129,000 verified software vulnerabilities through Project Glasswing partners and open-source efforts. Of these, more than 33,000 were rated critical- or high-severity.

Organizations interested in applying for CVP access can do so via Anthropic’s portal. The program is available on the Claude Platform, Google Cloud’s Vertex AI, and Microsoft Foundry, while Amazon Bedrock access is limited to customers eligible for specific safeguards.