TLDR Tech / Launches
Anthropic Offers Reduced AI Safety Controls for Security Teams
Anthropic is expanding its Cyber Verification Program with three access tiers that feature diminished AI safety restrictions for authorized security teams. These tiers are designed for defensive operations, approved red teaming, and specialized testing of critical infrastructure. According to Anthropic's tests using CyScenarioBench, these varying access levels significantly impacted the success rates of models in complex cyber operations. Industry observers emphasize the need for external oversight, strict authorization, and ongoing monitoring of such capabilities.
Read the full story