Associated Press reporter Garance Burke reported in a story published by ABC News on 8 November that a small AI safety unit in the US administration lacks a permanent director. It has also stopped publishing its research.

The unit, the Center for Advancing Innovation and Standards for Super Intelligence (CAISSI), is an independent division of the Commerce Department. Its job is to assess the safety of new AI products. Until recently it was called the U.S. Center for AI Standards and Innovation. Super Intelligence is President Donald Trump’s preferred term for AI.

Report based on anonymous sources

Two former administration officials told AP that CAISSI officials had recently been asked to report informally to a White House office. The two spoke on condition of anonymity to avoid repercussions for current staff. AP did not say who made the request.

The White House did not respond to AP’s detailed questions about CAISSI. In a statement, it said Trump had formally established a Super Intelligence Force and commissioned a report on AI risks from his science and technology council. Director of National Intelligence Jay Clayton leads the Super Intelligence Force.

The director the centre hired in April was fired within days, and the successor resigned after a few months. The director of the centre’s parent institute now leads it on an interim basis. CAISSI has an annual budget of $10 million, about one-eighth of the budget of its British counterpart, AP reported.

In August, Congress began asking why the centre had stopped publishing reports on the safety of certain AI systems. By then OpenAI had already asked policymakers to give the centre more money and support.

Critics fear findings will be suppressed

Critics fear the executive branch may politicise decisions that could affect areas such as cybersecurity and biological weapons. They say the centre’s findings could be altered or suppressed to benefit the technology industry. Others think closeness to Trump could bring the centre visibility and perhaps more money.

Amos Toh, senior counsel at the Brennan Center for Justice, said the administration is framing model safety as a national security matter that requires executive discretion and secrecy. Toh said this overshadows risks to consumer protection, privacy and civil rights.

In September, Trump asked leading AI companies to sign up to a voluntary safety pact and ‘tremendous self-policing’. The administration has not said how it will monitor implementation of the pact.

Logan Koepke of Upturn, a technology and civil rights organisation, questioned the role of government testing if the largest companies are encouraged to build their own evaluations or hire outside evaluators. The government’s ability to inspect AI systems independently has not kept pace with developments, AP reported. Sayash Kapoor, an incoming assistant professor at the University of California, Berkeley, said there is still no established way to test the safety of AI agents – programs that use a model to act autonomously – in real-world environments.