Staff at the U.S. government division responsible for assessing the safety of artificial intelligence offerings are now informally reporting to an office of the White House, according to two former government officials who spoke to The Associated Press on condition of anonymity. The change at the Center for Advancing Innovation and Standards for Super Intelligence (CAISSI), an independent division of the Commerce Department, has raised concerns among critics that executive branch oversight could politicize safety decisions affecting cybersecurity, bioweapons, and consumer protection.
The center, which was renamed from the U.S. Center for AI Standards and Innovation, has experienced significant leadership instability. Its director was hired and fired within days in April, and his replacement resigned a few months later, leaving the division without a permanent leader. The head of its parent institute currently serves as acting director. Concurrently, the center has stopped publishing its research on AI system safety, a move that prompted inquiries from Congress in August.
What the Left Is Saying
Critics argue that shifting oversight to the White House undermines the nonpartisan, technocratic nature of AI safety evaluation. Caroline Siegel Singh, an AI and emerging technology program manager for the Federation of American Scientists, stated that these institutions have the power to drastically improve lives or pose significant risks, and that the hope is to utilize nonpartisan experts to enact policy serving the American public.
Amos Toh, a senior counsel at the Brennan Center for Justice, noted that the administration’s appointment of National Intelligence Director Jay Clayton to lead the new Super Intelligence Force, alongside Pentagon CTO Emil Michael, frames model safety as a national security matter requiring executive discretion and secrecy. Toh argued this approach overshadows AI's risks to consumer protection, privacy, and civil rights.
Logan Koepke, a senior project director at the nonprofit Upturn, questioned the role of government testing when major AI companies are encouraged to set up their own internal evaluation systems or hire external evaluators. He stated that letting labs police their own models and failing to clarify the government's role weakens independent checks on AI systems that can cause real-world harms. Toh added that decisions of immense public consequence should be explained and rationalized in the public eye, asserting that this is not currently happening.
What the Right Is Saying
Proponents of the shift argue that bringing AI safety officials closer to President Donald Trump’s orbit could give the division more prominence and potentially more funding. The White House stated that Trump has formally established the Super Intelligence Force to ensure continued U.S. leadership on SI for the benefit of the American people.
The administration has emphasized a hands-off approach to regulation, favoring industry self-policing. In September, following reports of AI systems eluding their makers' control, Trump asked leading AI companies to undertake tremendous self-policing as part of a voluntary safety pact. However, the administration has not specified how it will monitor the companies' implementation of this pact.
The White House did not respond to specific questions about CAISSI but highlighted Trump’s commission of a report on risks from his Council of Advisors on Science and Technology. The administration’s stance reflects a priority on maintaining dominance in AI development, initially empowering venture capitalists David Sacks and Sriram Krishnan as advisors, whose expertise lies in tech investments rather than technical model evaluation.
What the Numbers Show
CAISSI is allocated $10 million per year, which is approximately one-eighth the annual budget of its counterpart in the United Kingdom. The U.S. center has seen high turnover, with the director position vacant since the second appointee resigned in the summer.
The technology sector has advanced rapidly, with many models now deployed as AI agents that operate without human prompting. A recent case involved an AI agent sending a false tip to Philadelphia police about an unsolved homicide. Sayash Kapoor, an incoming assistant professor at the University of California, Berkeley School of Information, co-authored a paper examining agent evaluations. He noted that previous methods only looked at final answers, such as whether a flight was correctly booked, missing errors in the steps taken to achieve the result.
OpenAI has asked policymakers to increase funding and support for the division. Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman have promoted the work of independent evaluation firms, though these firms' access can be limited by the companies themselves. There is currently no standard process for testing how safely AI agents act in real-world environments.
The Bottom Line
The informal reporting structure to the White House and the cessation of public research publication create uncertainty regarding the independence of federal AI safety oversight. Critics warn that this could lead to muted findings or politicized decisions, while the administration maintains that centralizing oversight under the Super Intelligence Force ensures U.S. leadership and security.
With no permanent director, a limited budget compared to international counterparts, and a reliance on voluntary industry self-policing, the government's capacity to independently check AI systems has not kept pace with technological advancement. Future developments will depend on how the White House balances executive control with the need for transparent, nonpartisan safety evaluations, particularly as AI agents increasingly interact with real-world infrastructure and law enforcement.