The administration finished voluntary tests that gauge how capable frontier models are at cyberattacks as OpenAI, Anthropic, Google…