Anthropic Hands Cyber Weapons to Defenders

🔥 What happened Anthropic merged Project Glasswing and its Cyber Verification Program into a three-tier access system. Security teams can now apply for Defense, Red Team, or Specialized Access and get relaxed safety filters on Claude Opus 5.5, Sonnet 5.5, and Mythos 5.1. 💡 Why it matters On CyScenarioBench, the standard model blocked every cyber task at the first prompt. Defense Access let 4 of 50 trials through; Red Team Access saw zero blocks. Vetted defenders now wield capabilities ordinary users will never touch. ⚡ Our take Anthropic is building a two-class AI: trusted insiders get to practice offense, everyone else stays locked out. Pass the vetting, and you're playing a different game.