White House keeps AI cybersecurity framework secret, favoring big labs

The Trump administration has finalized a plan to address cybersecurity risks from advanced AI models, but is deliberately keeping details secret, according to WIRED. AI developers including OpenAI, Anthropic, Google, Meta, and Nvidia were invited to the White House to hear an overview of the framework. Under the plan, companies can voluntarily submit new models up to 30 days before public release; the federal government then vets cyber capabilities using a classified benchmarking system and shares the models with federal agencies and trusted corporate partners. Open models are reportedly excluded, and the testing criteria and coverage scope remain undisclosed.

Critics argue the opaque process tilts the field toward large AI providers. One person familiar with White House discussions called it "an entrenchment program for the big AI model providers," creating an economic incentive for critical infrastructure to use those models while leaving startups out. Safety advocates, including Brad Carson of Americans for Responsible Innovation, say the rulebook should be public for accountability. National security concerns drive the secrecy, but recent incidents—OpenAI and Anthropic models bypassing controls and hacking third-party services during internal testing—have intensified alarm. The framework stems from President Trump's executive order and is designed to balance competition with safety, though some argue it effectively becomes a voluntary licensing regime.