Anthropic creates tiered AI access for verified cyber defenders
AI Cybersecurity
Anthropic creates tiered AI access for verified cyber defenders
AnthropicOct 7, 2026
Anthropic introduced three access tiers for vetted security professionals, offering differentiated access to its models for defensive work and authorized testing.
Anthropic announced an expanded Cyber Verification Program on October 6, combining and extending access previously offered through its Cyber Verification Program and Project Glasswing. The program gives verified security professionals access to advanced cyber capabilities and less restrictive safeguards on models including Claude Opus 5.5, Sonnet 5.5 and Mythos 5.1. Anthropic says the aim is to help defenders find and address vulnerabilities while restricting access that could enable misuse.
The updated program has three tiers. Defense Access covers work such as incident response, malware analysis and validating vulnerabilities in systems an organization owns or maintains. Red Team Access adds authorized penetration testing and adversarial assessments. Specialized Access, which Anthropic says is reserved for a limited set of verified organizations, covers testing of sensitive systems such as power grids, telecom networks and government networks. The company says higher-risk work requires deeper verification, and that some harmful actions remain blocked even for red-team users.
Anthropic published its own evaluation of how the tiers affect model behavior on CyScenarioBench. It says the model was blocked on all tested challenges without program access; in Defense Access, safeguards blocked 46 of 50 trials; in Red Team Access, the model completed 34 of 50 tasks. Those figures are Anthropic’s results on a particular evaluation, not a guarantee that the controls prevent misuse in every setting. The company also says program participants must meet security requirements and that data retention is required for monitoring, with a separate zero-data-retention option planned. The announcement reflects a broader challenge for AI security: capabilities that help defenders audit systems can also assist attackers, so access controls and verification become part of the product itself.
Industry Impact
Controlled access for cyber operations
Cybersecurity, critical infrastructure, telecommunications, and government networks
Overview
Anthropic’s three-tier program gives vetted security professionals access to advanced model capabilities for defensive work and authorized testing, with deeper verification for higher-risk activity. Security teams may gain new support for incident response, malware analysis, and vulnerability testing, while organizations handling sensitive systems face additional access and data-retention requirements. Anthropic’s CyScenarioBench results illustrate how safeguards vary by tier, but reflect one evaluation rather than a guarantee against misuse. The model could influence how enterprises balance security-team productivity with controls on powerful AI.
Key Impact Points
Security professionals can apply for differentiated access aligned with defensive, red-team, or specialized testing needs.
Organizations testing sensitive infrastructure must account for deeper verification and required data retention.
Anthropic’s benchmark results show substantial differences between access tiers, but do not establish universal protection against misuse.
Opportunities
Apply AI to authorized security workflows
Enterprise cybersecurity and security testing
Verified organizations may be able to use the program to support incident response, malware analysis, vulnerability validation, and authorized penetration testing. The practical opportunity is to match defined security tasks to the appropriate access tier, then integrate model use into existing review and authorization processes. Organizations should first confirm eligibility and understand the program’s security and data-retention requirements; Anthropic says a separate zero-data-retention option is planned. The announcement does not disclose pricing or a direct revenue model, so potential value is operational—supporting defensive work and testing—rather than a stated new revenue stream.
Tiered AI support for authorized security testing
Problem: Security teams need capable tools for defensive analysis and authorized testing, while access to cyber capabilities can also create misuse risks.
Solution: Eligible organizations can seek access through Anthropic’s tiers, which distinguish defensive work, authorized red teaming, and specialized testing of sensitive systems.
Target Users: Verified security professionals and organizations responsible for systems they own, maintain, or are authorized to test.
Business Value: Potentially supports vulnerability validation, incident response, malware analysis, and adversarial assessments within a verified-access program.
Revenue Potential: No pricing or direct revenue model is disclosed; the opportunity is operational value for participating organizations.
