AI safety center loses independence as White House takes reporting role

The U.S. government unit responsible for testing the safety of new artificial intelligence models now reports informally to the White House. This shift occurs while the agency lacks a permanent director and has stopped publishing its research findings to the public.

Two former government officials with knowledge of internal discussions told The Associated Press that staff at the Center for Advancing Innovation and Standards for Super Intelligence were recently asked to adopt this new reporting structure. The division, known by the acronym CAISSI, operates as an independent branch of the Commerce Department but faces scrutiny over its autonomy.

Critics argue that placing the center under executive branch influence risks politicizing technical assessments. They worry that findings regarding cybersecurity and bioweapons risks could be altered or muted to appease the tech industry. Caroline Siegel Singh, a program manager at the Federation of American Scientists, stated that the goal should be nonpartisan technocratic experts serving the American public. Others contend that closer proximity to President Donald Trump could increase the division's prominence and funding.

The center has faced leadership instability since its reorganization. A director hired in April was fired within days. His successor resigned after a few months. The position has been vacant since then, with the head of the parent institute serving as acting director. The agency receives an annual allocation of $10 million, which is roughly one-eighth of the budget for the similar agency in the United Kingdom.

Congress began questioning the halt in public reports in August. By that time, OpenAI had urged policymakers to provide more financial support to the division. The White House did not answer specific questions about CAISSI. It issued a statement confirming that President Trump established his Super Intelligence Force. The statement said the president commissioned a report on risks from his Council of Advisors on Science and Technology to ensure continued U.S. leadership in the field.

The approach to AI safety differs from other industries. The National Highway Traffic Safety Administration conducts public crash tests to set vehicle standards. Under former President Joe Biden, AI companies voluntarily agreed to share details with the Commerce Division. In September, President Trump asked leading AI firms to undertake tremendous self-policing under a voluntary safety pact. His administration has not specified how it will monitor these companies' implementation of the pact.

Technology advances have outpaced government testing capabilities. Many models now function as AI agents that plan steps in real-world environments without human prompting. An AI agent recently sent a false tip to Philadelphia police regarding an unsolved homicide. Sayash Kapoor, an incoming assistant professor at the University of California, Berkeley, noted that current methods often only check the final answer. He said that without reviewing every step an agent takes, evaluators cannot detect major errors in the process.

President Trump initially appointed venture capitalists David Sacks and Sriram Krishnan to lead AI policy. Both stepped down earlier this year. Sacks now serves on the presidential technology advisory council. Following a resignation by an Anthropic engineer who warned about development dangers, Trump empowered National Intelligence Director Jay Clayton to lead the Super Intelligence Force. Amos Toh, a senior counsel at the Brennan Center for Justice, said this frames safety as a national security matter requiring executive discretion.

Silicon Valley firms have disclosed their models' roles in cyberattacks this year. CAISSI has maintained a low profile during this period. Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman are promoting independent evaluation firms. However, these companies can limit access to external evaluators. Logan Koepke, a director at the nonprofit Upturn, asked what role government testing plays if large companies hire their own external evaluators. Toh stated that these decisions have immense public consequence and should be rationalized in the public eye.

The White House statement emphasized that the new structure would benefit the American people. CAISSI remains without a permanent leader as it adjusts to the new reporting lines. The next scheduled step involves the completion of the report commissioned by the Council of Advisors on Science and Technology.