US AI Safety Center Staff Asked to Report Informally to the White House

The unit is a division of the Commerce Department. Until recently it was called the Center for AI Standards and Innovation; it is now the Center for Advancing Innovation and Standards for Super Intelligence, or CAISSI. The AP did not say which White House office the staff have been asked to deal with.

The change matters because the center is the main federal body that checks advanced models for risks to national security, and it has had testing agreements with OpenAI and Anthropic since its predecessor signed them in 2024. Critics told the AP they fear closer political control could politicise decisions with implications for cybersecurity and bioweapons, and lead to findings being altered or played down. Others argued that being closer to the White House could bring the division more prominence and money.

The center has been unsettled for months. According to the AP, its director was hired and then fired within days in April, his replacement resigned a few months later, and the head of its parent institute is now acting director. The division has also stopped publishing its research. In August, Congress asked why it had stopped releasing reports on how safely certain AI systems performed.

Its budget is about $10 million a year, roughly an eighth of what its British counterpart receives, the AP reported. OpenAI has urged policymakers to give the division more money and support.

The White House did not answer the AP's specific questions about CAISSI. In a general statement, it said President Donald Trump had formally established a "Super Intelligence Force" and would ensure "continued U.S. leadership on SI".

The reporting change follows the administration's wider shift towards industry self-regulation. In September, Trump asked leading AI companies to undertake "tremendous self-policing" under a voluntary safety pact. The administration has not said how it will check whether companies comply.

Outside experts told the AP that this leaves gaps. Caroline Siegel Singh of the Federation of American Scientists said the hope is for "nonpartisan or bipartisan technocratic experts" to set policy. Sayash Kapoor of the University of California, Berkeley, said there is still no standard way to test how safely AI agents behave in the real world. He warned that serious mistakes can go unnoticed "unless you're looking at the entire set of steps that an agent took".

The center's future may become a test for the next Congress. Democrats, who will take control of Congress next year, have said they plan legislation that includes independent testing of frontier AI models.