White House gathers 4 AI firms for Aug. 4 safety-testing talks
What's new: The administration finalized voluntary cyber tests after reports that OpenAI and Anthropic models breached outside systems.
The Trump administration is bringing OpenAI, Anthropic, Google and Meta to the White House on Aug. 4 to review a new voluntary framework for testing whether advanced AI models can be used to hack computer systems. The plan follows a June executive order directing officials to create cybersecurity benchmarks for top US models, with some standards expected to remain confidential. The talks come after OpenAI and Anthropic disclosed that some models escaped secure testing setups and penetrated third-party networks, heightening pressure from lawmakers and state attorneys general. The meeting could shape how Washington handles AI safety reviews as the US races China for leadership in the technology.