Staff at Commerce’s AI Testing Center CAISSI Were Asked to Report Informally to the White House, AP Reports
Source: The Associated Press (via Tech Times) · published November 8, 2026
Staff at the federal government’s main unit for testing frontier AI models were recently asked to report informally to a White House office, The Associated Press reported on November 8, citing two former officials who spoke on condition of anonymity. The unit, the Commerce Department’s Center for Advancing Innovation and Standards for Super Intelligence (CAISSI), grew out of the Biden-era U.S. AI Safety Institute, and critics told the AP the arrangement raises questions about its independence.
Tech Times, summarizing the AP report, said the arrangement is informal rather than a formal transfer. It follows an October report by The Information, relayed by Reuters, that the administration planned to move three senior CAISSI officials into the White House to report to Office of Science and Technology Policy Director Michael Kratsios and National Cyber Director Sean Cairncross. The center has had no permanent director since Chris Fall resigned in July, and the AP says it has stopped publishing its research. NIST’s public list shows assessments of Chinese models and joint guidance, but no published assessment of a U.S. developer’s model.
Executive Order 14409, which governs federal pre-release testing of AI models, does not mention CAISSI and contains no requirement to publish assessments, according to Tech Times. Amos Toh of the Brennan Center for Justice and Logan Koepke of Upturn told the AP that decisions touching cybersecurity and bioweapons risks could be politicized, and questioned what role government testing plays if labs run their own evaluations. Some observers argue closer ties to the White House could bring the division more prominence and possibly more money; the AP put its budget at about $10 million a year, roughly one-eighth of its U.K. counterpart’s.
What it means in practice
Sayash Kapoor, an incoming assistant professor at UC Berkeley’s School of Information, told the AP that older evaluation methods check only a model’s final result, so a major mistake by an AI agent can go unseen unless evaluators examine every step it took. There is still no standard process for testing how safely agents act in the real world. The report adds to questions about federal oversight that followed the White House order mandating AI incident reporting and House Democrats’ request that frontier labs report safety incidents within 48 hours.
What to watch is whether Commerce names a permanent CAISSI director, whether the center resumes publishing assessments of U.S. frontier models, and whether the incoming Congress uses its oversight power to press for a documented reporting line and public results.
Read the original report: The Associated Press (via Tech Times)