Left and right unite over secret White House AI rules

Campaigners, consumer groups and free-market advocates warn about concentration of oversight power

A diverse coalition of advocacy groups, industry representatives and non-profits has written to President Trump asking the White House to lift the veil of secrecy over its AI review framework.

The framework reportedly focuses on the most capable proprietary AI models developed by US companies, particularly those that demonstrate advanced cybersecurity or hacking capabilities. Developers of these models can voluntarily submit their systems for government testing before public release, giving the authorities a first look.

Signatories to the open letter say they support the administration’s emphasis on AI innovation and safety, but are concerned about the secrecy surrounding the federal government’s framework for reviewing new frontier models. They argue that the public, Congress, researchers and smaller AI developers also need to understand how the system works.

The signatories represent a broad mix of political opinion, from progressive civil rights organisations such as the Center for American Progress to libertarian groups like Americans for Prosperity, tech policy advocates including the Center for Democracy and Technology, and transparency and rights campaigners such as Government Information Watch.

Their shared concern is that the federal government and a handful of companies could shape the future direction of a powerful technology unless there is proper scrutiny.

The oversight framework was reportedly shared in August with representatives from OpenAI, Anthropic, Meta, Google, Nvidia and other leading AI developers, but it has not been made public.

On transparency and accountability, the signatories recognise the need to keep some details secret for security reasons, but they argue that “the government has an unequivocal duty to disclose information about the framework’s standards, requirements, structure, and operational details.”

A lack of public oversight of the testing framework could give the federal government undue power to deny or restrict access to “the most transformative technologies we have ever seen,” they say.

Similarly, if the rules of the game are only divulged to the largest frontier labs, smaller developers will be disadvantaged and incumbency entrenched.

Restricting Congress's access to the framework could impede its efforts to develop policy and guardrails for frontier AI, they argue.

Finally, the letter suggests, secret governance arrangements could undermine public trust.

The letter is supportive of the US government’s efforts to maintain its lead over China, but argues that public participation will be needed to further this goal in the future.

“With more public discourse we can ensure the government is successful in its goal to maintain its lead over China in the AI race and to advance innovative and safe AI frontier models,” it concludes.

The letter comes amid reports that Anthropic, which markets itself as the most responsible big AI lab, excluded the UK’s AI Security Institute from reviewing its latest Claude Mythos 5.1 model before launch, breaking a precedent that had existed since 2023 following the AI Safety Summit, when the UK positioned itself as a global leader in AI safety.

Together, they suggest that scrutiny of the world's most powerful AI systems may increasingly be confined to a small circle of US government agencies and frontier labs.

"Anthropic failing to submit its latest model to Britain's AI Security Institute for testing highlights a strategic exposure in the UK and Europe," commented Olev Volpin, group president at Amdocs International.

"Companies, including critical infrastructure, are reliant on end-to-end foreign value chains. Yet regulators are not moving fast enough to plan for the possibility that US companies cut off their services outside the border.

“It is not yet confirmed whether Anthropic's decision came as a request from the government, but it shows a reluctance to share knowledge about their latest models.