AUTO-UPDATED

Anthropic loosens Claude’s cyber restrictions for verified defenders

Anthropic has expanded its Cyber Verification Program to grant verified security professionals access to advanced Claude models with fewer restrictions for defensive tasks and authorized penetration testing.

Key Points

  • The program features three tiers—Defense Access, Red Team Access, and Specialized Access—tailored to specific cybersecurity workflows and risk levels.
  • Eligible participants include security teams at businesses, government agencies, and critical infrastructure operators, with verification required for higher-tier access.
  • Anthropic tested Claude Opus 5.5 using CyScenarioBench, finding that Red Team Access allowed the model to complete 34 of 50 multistage cyber operation trials.
  • Project Glasswing partners utilized Claude Mythos models to identify over 129,000 verified software vulnerabilities between April and July 2026.
  • The program is available via the Claude Platform, Google Cloud Vertex AI, and Microsoft Foundry, with Amazon Bedrock access limited to specific enterprise customers.

Why it Matters

This initiative balances the need for powerful AI tools in cybersecurity with strict safety guardrails to prevent the misuse of models for malicious activities. By formalizing access for verified professionals, Anthropic enables more efficient vulnerability detection while maintaining oversight of high-risk operations.
Help Net Security Published by Anamarija Pogorelec
Read original