Commerce Division CAISSI to Report Informally to White House Amid Leadership Vacuum
The AI safety unit, which has no permanent director and has paused public research, is now linked to the executive branch, raising questions about independence.
By Riya Zhao ·
The Center for Advancing Innovation and Standards for Super Intelligence (CAISSI) has been asked to report informally to the White House, a move that two former government officials say complicates the independence of the federal unit tasked with assessing artificial intelligence risks.
The division, a part of the Commerce Department, currently lacks a permanent leader. Its director was hired and fired within days in April. His replacement resigned a few months later, leaving the head of the parent institute to serve as acting director. The center has also stopped publishing its research on how safely specific AI systems are performing, a pause that prompted inquiries from Congress in August.
Critics argue the new reporting structure allows the executive branch to influence decisions affecting cybersecurity and bioweapons. Caroline Siegel Singh, a program manager at the Federation of American Scientists, stated that the goal is to use nonpartisan experts to enact policy that serves the American public. She noted that these powerful institutions can drastically improve lives or produce the opposite result.
CAISSI receives an annual allocation of $10 million. This figure represents approximately one-eighth of the annual budget for the United Kingdom's counterpart, an agency that has a lower profile in frontier technology development. In April, OpenAI urged policymakers to provide the division with additional financial support and resources.
The administration has shifted the focus of AI oversight toward national security. President Donald Trump established a Super Intelligence Force led by National Intelligence Director Jay Clayton. The White House said in a statement that Trump commissioned a report on risks from his Council of Advisors on Science and Technology to ensure continued U.S. leadership in the field.
Industry experts question the current testing framework. Sayash Kapoor, an incoming assistant professor at the University of California, Berkeley, noted that previous methods often only checked the final output of an AI task. He explained that without reviewing the entire set of steps an agent takes, evaluators cannot detect if the system made a major error along the way. This is a concern as AI agents now operate without human prompting in real-world environments.
Amos Toh, a senior counsel at the Brennan Center for Justice, said the administration is framing model safety as a national security matter requiring executive discretion. He argued that this approach overshadows risks to consumer protection, privacy, and civil rights. Toh stated that decisions with immense public consequence should be explained in the public eye, a process he said is not currently happening.
