By Hamza Faisal

The small US government office charged with testing advanced artificial intelligence for safety has no permanent leader and has stopped publishing its research, according to the Associated Press — and its staff have recently been asked to report informally to a White House office.

The Center for Advancing Innovation and Standards for Super Intelligence, known as CAISSI, is an independent division of the Commerce Department, housed within the National Institute of Standards and Technology (NIST). Its work is to evaluate frontier AI systems and publish findings that help government agencies and companies understand the risks of the most powerful models.

The arrangement has raised questions among outside experts about the independence of government AI evaluations, at a moment when public anxiety is growing over reports of AI agents behaving unpredictably.

A revolving door at the top

CAISSI has struggled to keep a director. Collin Burns, a former Anthropic researcher, was pushed out of the role after just four days in April, reportedly over his links to the company. His replacement lasted only a few months before resigning, and the centre has been without a permanent leader since, with Arvind Raman, the head of NIST, serving as acting director.

In August, Congress began asking why the centre had suddenly stopped making its reports on AI systems' safety public. The centre is allocated $10 million a year — about one-eighth of the budget of its UK counterpart — and OpenAI has publicly called for more support for it.

Reporting to the White House

Two former government officials told the Associated Press that CAISSI staff have recently been asked to report informally to an office of the White House. No formal transfer of the centre has been announced, and the exact scope of the arrangement is unclear.

The White House did not answer the news agency's specific questions about CAISSI. Instead, it pointed to President Donald Trump's new Super Intelligence Force, which is led by National Intelligence Director Jay Clayton. In September, Trump called for AI companies to undertake "tremendous self-policing" through a voluntary safety pact, but the administration has not explained how it will monitor compliance.

What is still unclear

The picture is not entirely blank. NIST's website lists work the centre released in September and October, including a September 17 assessment of an AI model's cyber capabilities and incident-response guidance published with other agencies on October 16. Those releases do not resolve which research the Associated Press describes as halted, or when the halt began.

Until recently the centre was known as the US Center for AI Standards and Innovation. Its evaluations are meant to be a neutral check on systems built by the same companies racing to deploy them. With no permanent director, no public reports, and an informal line to the White House, the question now is who is watching the watchers — and when the public will see their findings again.