America’s federal AI safety center has stopped publishing research and remains without a permanent director as its staff begin informally reporting to the White House. The Center for Advancing Innovation and Standards for Super Intelligence, housed within the Commerce Department, assesses the safety of increasingly capable AI systems.

Two former officials told The Associated Press about the reporting arrangement, raising questions about its independence from an administration promoting rapid technological development. Leadership has been unstable: one director was hired and dismissed within days, and his successor resigned after several months. Congress began questioning the disappearance of public safety reports in August.

The center receives $10 million annually, roughly one-eighth of its British counterpart’s budget. Supporters of closer White House involvement suggest greater access could bring resources and influence, while critics worry that unfavorable findings could instead be suppressed or softened.

The White House has not addressed these new developments about the center, instead pointing to Trump’s new Super Intelligence Force and a commissioned risk report while emphasizing American leadership.

Independent testing addresses both a commercial problem and a regulatory one. Businesses buying AI cannot easily inspect how a model behaves across thousands of situations, while developers have incentives to emphasize capability and speed. Public evaluations provide customers with evidence beyond the supplier’s assurances…

Elon Musk is on record predicting this new opportunity is set to be 180 times bigger than the SpaceX IPO.

That IPO minted 4,400 millionaires overnight.

If this is 180 times bigger…

Just imagine how many more Americans will become millionaires.

Will you be one of them? Click here now.

AI agents introduce a problem that conventional product tests can overlook. For example, booking the right flight counts as a success, but not if the software exposes payment details in the process. Evaluators need to examine how a system completes its task, since a correct result can conceal costly mistakes.

Published government research could help businesses identify those weaknesses before entrusting software with customer records or financial transactions. Without it, buyers have less evidence against which to assess developers’ claims, while outside researchers cannot examine the government’s findings or challenge its methods.

White House involvement should at least ostensibly provide greater funding and influence, while also placing evaluators closer to officials whose priority is accelerating American AI development. Findings that warrant delaying a release may conflict with that ambition.

When safety findings disappear from public view, Americans lose a way to judge the risks companies ask them to accept. Employers purchasing AI may know its price and advertised capabilities without knowing how readily it exposes confidential records or follows malicious instructions. Such weaknesses can become customers’ problems before buyers discover them.

Small businesses may find themselves particularly disadvantaged. Larger companies can commission evaluations or employ specialists, while smaller firms lack the resources and may consequently find themselves relying more heavily on suppliers’ assurances.

While the administration’s pursuit of American leadership is understandable, suppressing evidence of weaknesses ultimately leaves American businesses less equipped to deploy the technology successfully. Faster adoption brings little competitive advantage if firms cannot trust the systems they purchase. Giving evaluators greater access to the White House could strengthen their influence, while keeping their work public would allow that influence to reach the businesses expected to turn America’s technological lead into economic gains.