Two former government officials told the Associated Press on Sunday that staff at the U.S. government’s Center for Advancing Innovation and Standards for Super Intelligence (CAISSI) had recently been asked to report informally to a White House office. The officials spoke anonymously, and the White House did not answer AP’s specific questions about the arrangement. The report describes an informal request, not a publicly announced reorganization.
The distinction matters because CAISSI is one of the few federal units tasked with evaluating advanced AI systems. The National Institute of Standards and Technology says the center works on testing, voluntary standards and unclassified assessments of risks including cybersecurity, biosecurity and chemical weapons. Those functions make its reporting chain part of the substance: who sets evaluation priorities can shape which risks are studied and what results reach the public.
A testing center with a thin public record
AP also reports that CAISSI has no permanent director and has stopped publishing its research. The center’s public NIST page lists an October 16 update on incident-readiness practices and several earlier model assessments, but a public page alone cannot show whether unpublished work continues internally. Congress began asking in August why safety-assessment reports were no longer being released, according to AP.
The reporting gap is consequential even without evidence that any result has been changed. AI developers often conduct their own tests, while outside evaluators may depend on a company’s permission and access. A government center can offer a different kind of check, particularly when its methods and findings are visible enough for researchers, policymakers and the public to scrutinize. If that work becomes less transparent, outsiders have fewer ways to tell whether a model’s risks were tested consistently.
What the report does—and does not—establish
The AP account relies on two former officials familiar with the discussions. The White House’s response, as quoted by AP, described President Donald Trump’s new “Super Intelligence Force” and a risk report commissioned from his science and technology advisers; it did not directly confirm or deny the informal reporting request. That leaves an important gap: the public does not yet know which office requested the change, whether it is temporary, or whether CAISSI’s technical decisions and publication authority have shifted.
There is also a plausible administrative case for closer coordination. CAISSI’s mandate already calls for work with other federal agencies, and a White House link could make it easier to coordinate across departments or raise the center’s profile. AP quoted supporters who said closer ties might bring prominence or funding. Those possibilities do not answer the independence question, but they caution against treating a reporting request by itself as proof that safety assessments have been politicized.
TINA’s view
The best-supported conclusion is narrower than either alarm or reassurance: the oversight picture is unsettled. A small evaluation unit with leadership turnover and fewer public research releases is difficult to assess from the outside; an informal line to the White House adds another unanswered question. The available reporting does not show that officials suppressed a finding, directed a test, or altered a result. It does show why independent, reproducible evaluation matters as AI systems take on tasks with consequences beyond a chat window.
Evidence that would change this judgment is concrete: a published memorandum spelling out the reporting relationship, a named permanent director, and a regular schedule for releasing methods and assessments. If the arrangement is only administrative and the center continues to publish independent work, that would ease the concern. If reports remain unavailable while decisions move behind closed doors, readers will have stronger grounds to ask who is checking the checkers.
For now, watch CAISSI’s public NIST updates and any formal statement from Commerce or the White House. The next useful signal is not another assurance about AI leadership; it is a clear account of who directs the evaluations and whether their results remain public.
