The government centre responsible for evaluating advanced artificial intelligence lacks a permanent leader and has stopped publishing assessments

SAN FRANCISCO — Employees at the U.S. government centre charged with testing advanced artificial-intelligence systems have been asked to report informally to a White House office, according to two former officials cited by The Associated Press.

The reported arrangement is drawing scrutiny because the Center for Advancing Innovation and Standards for Super Intelligence, or CAISSI, is intended to provide technical expertise inside the U.S. Commerce Department.

The centre has no permanent director and has stopped publishing its model evaluations, AP reported Sunday. Members of Congress began asking in August why those reports had ceased.

AP based its account of the reporting arrangement on two unnamed former officials familiar with the discussions. The White House did not answer the news agency’s specific questions about CAISSI.

A small office with a large mandate

CAISSI is part of the National Institute of Standards and Technology. Its official mandate includes developing voluntary security standards and conducting unclassified evaluations of AI capabilities that could pose cybersecurity, biological or chemical risks.

Its published evaluations have included tests of Chinese-developed, open-weight AI models and comparisons with American systems.

AP reported that the centre receives about US$10 million annually, roughly one-eighth of the budget available to its British counterpart. It has been without a permanent leader since a series of departures earlier this year.

The White House told AP that President Donald Trump had established a Super Intelligence Force and commissioned a risk report from the President’s Council of Advisors on Science and Technology. It did not explain how CAISSI would operate within that structure.

Independence versus influence

Moving the centre closer to the White House could increase its visibility and funding. It could also make technical findings more vulnerable to political or national-security restrictions.

That distinction matters because the government is increasingly relying on voluntary commitments from AI developers. Trump asked major companies in September to undertake extensive self-policing, but the administration has not publicly explained how compliance will be independently measured.

AI agents create a particular testing challenge. Unlike a conventional chatbot, an agent may browse websites, use software, handle credentials or execute a long sequence of actions. A test that examines only the final result can miss unsafe or unauthorized steps taken along the way.

Relevance beyond the United States

American AI evaluations influence procurement, international standards and policy discussions in Canada. Canadian institutions also use models produced by the companies that CAISSI is supposed to assess.

A loss of public testing would leave governments, businesses and researchers more dependent on safety reports produced or commissioned by AI companies themselves.

The immediate questions are whether the White House will formalize the reported reporting relationship, when a permanent director will be appointed and whether CAISSI will resume publishing detailed evaluations. Until those points are clarified, concerns about political control remain grounded in AP’s reporting rather than a publicly released organizational directive,