Anthropic expands its Cyber Verification Program by integrating Project Glasswing and offering three tiers, all with access to its most capable Claude models

Access to Anthropic's most capable AI models for cybersecurity tasks is now tiered, meaning more security professionals can bypass some AI safety blocks for their work.

Anthropic has launched an expanded Cyber Verification Program (CVP), integrating its Project Glasswing and offering three distinct access tiers for security professionals. The program now provides qualifying individuals and organizations with enhanced access to Anthropic's most capable AI models, including Claude Opus 5.5, Claude Sonnet 5.5, and Claude Mythos 5.1. The tiers are Defense Access for general security operations and incident response, Red Team Access for authorized penetration testing, and Specialized Access for highly sensitive critical infrastructure testing. Each tier has varying verification requirements and security controls, with access to reduced blocking classifiers tailored to specific cybersecurity work. The program aims to balance the dual-use nature of AI by limiting harmful activities for general users while empowering security professionals with advanced tools. Anthropic conducted evaluations using CyScenarioBench, demonstrating significant reductions in model blocks across the tiers, with Specialized Access showing no blocks and a high task completion rate.

Anthropic's strategic integration of Project Glasswing into its Cyber Verification Program signifies a deliberate move to broaden the availability of advanced AI capabilities for cybersecurity defenders. By establishing tiered access, Anthropic is segmenting the market for AI-assisted security operations, directly addressing the need for specialized tools that can navigate the complexities of modern cyber threats without overly restrictive safeguards. This expansion is particularly impactful for organizations engaged in critical infrastructure protection and red-teaming, sectors that previously faced limitations due to the dual-use nature of AI.

The introduction of Defense, Red Team, and Specialized Access tiers represents a sophisticated approach to risk management for AI developers. It suggests a growing trend in the AI industry towards offering nuanced access controls that cater to specific industry verticals and professional roles, rather than a one-size-fits-all approach. This model allows Anthropic to foster innovation in cybersecurity while maintaining ethical boundaries, potentially setting a precedent for how other AI providers will balance capability with safety in sensitive applications.