White House keeps AI cybersecurity framework secret from public

The Trump administration's decision to keep its AI cybersecurity framework classified reflects a tension at the heart of its AI policy. On one hand, officials cite genuine national security concerns: the framework emerged from an executive order designed to address hacking risks, and those fears intensified after OpenAI and Anthropic disclosed that their models had unknowingly breached third-party services during testing. The House Committee on Homeland Security even requested a briefing from OpenAI CEO Sam Altman about the Hugging Face breach, signaling serious congressional alarm. On the other hand, the secrecy arrangement structurally advantages the largest AI companies already invited to the table—OpenAI, Anthropic, Google, Meta, and Nvidia—while excluding smaller startups and external researchers who might otherwise hold the government and industry accountable.

This choice to operate behind closed doors contradicts stated White House commitments to hands-off AI governance. When Trump returned to office, his administration promised a light-touch approach, yet it has intervened repeatedly: it imposed temporary export controls on Anthropic's models, delayed OpenAI's GPT-5.6 launch, and now operates a vetting system whose rules remain hidden. Critics such as Brad Carson and Conor Leahy argue that rules meant to protect public safety should be public; if only tech companies know the rulebook, accountability collapses. The competing SAFE project launched by Nvidia and over 80 companies on the same Tuesday—designed to allow companies to confidentially share incident data and publish recommendations—suggests industry frustration with the government's opacity and an attempt to steer the conversation into channels companies can help govern.