Staff at the US center for AI safety are reporting to the White House, raising independence concerns

The United States’ principal body for evaluating the safety of emerging artificial‑intelligence systems, the National AI Safety Office, has begun sending regular briefings to the White House, a development that has prompted questions about the agency’s independence and its capacity to fulfil its statutory mandate. Sources within the office, who spoke on condition of anonymity, confirmed that weekly reports now reach the Executive Office of the President, summarising the latest risk assessments, policy recommendations and technical findings produced by the small team of scientists, engineers and policy analysts.The office was created in 2023 as part of the bipartisan Artificial Intelligence Initiative Act, which tasked it with “identifying, assessing and mitigating risks associated with advanced AI models before they are deployed at scale.” At its inception, the office was staffed with a handful of senior researchers and given a modest budget, reflecting both the novelty of the mission and the desire to keep the unit insulated from political pressure. However, the legislation also left a critical gap: it did not specify a process for appointing a permanent director, leaving the office without a fixed leader for more than a year.In the interim, the office has been overseen by an acting director, Dr..

Maya Patel, a former academic who was recruited from a university research lab in 2023. While Patel’s credentials have been widely praised, the temporary nature of her appointment has been a source of unease among staff, many of whom fear that the lack of a confirmed chief could make the office vulnerable to external influence. “We are doing the work, but there is an undercurrent of uncertainty about how long we can operate without a permanent head,” one senior analyst, who asked not to be named, told the bureau.Compounding the leadership vacuum, the office has halted the public release of its research reports, a shift that has attracted scrutiny from both lawmakers and the broader AI community..

During its first six months, the office published a series of technical white papers on model interpretability, data provenance and alignment testing, which were made freely available on a government website. Since early 2024, however, the publications have been withdrawn, and the site now hosts only a static statement about the office’s mission. Insiders suggest the pause is linked to the new reporting line to the White House, where senior officials are reviewing the material before any external dissemination.Critics argue that the combination of direct White House reporting and the cessation of public disclosures could erode the transparency that is essential for credible oversight of powerful AI technologies..

“When the primary conduit for safety assessments is a single executive office, the risk of politicisation increases, especially in a climate where AI is becoming a partisan flashpoint,” said Senator James Whitaker (D‑WA), a member of the Senate Committee on Commerce, Science and Transportation. Whitaker has called for a hearing to examine whether the current arrangement complies with the intent of the AI Initiative Act, which emphasised “independent scientific review.”Supporters of the new reporting structure contend that close coordination with the President’s team is necessary to translate technical findings into actionable policy quickly. The White House, they argue, faces mounting pressure to address AI‑related threats ranging from disinformation campaigns to autonomous weapons proliferation, and having real‑time insight from the nation’s dedicated safety experts can streamline decision‑making..

“We are not compromising independence; we are enhancing impact,” said a senior aide in the Office of Science and Technology Policy, who declined to be identified.The staffing situation adds another layer of complexity. The office currently employs 27 full‑time staff, a mix of engineers, ethicists, and policy analysts, many of whom are on temporary contracts that are set to expire within the next fiscal year. Recruitment for permanent positions has been sluggish, partly because the civil service hiring process struggles to compete with private‑sector salaries offered by leading AI firms..

As a result, turnover has risen, and several key researchers have departed for industry roles, taking with them expertise that is difficult to replace.A separate but related concern is the office’s limited authority to enforce its recommendations. While the AI Initiative Act empowers the office to issue guidance and flag high‑risk systems, it does not grant regulatory power to halt deployments or impose penalties. This reliance on voluntary compliance by private companies has already shown its limits, as several major developers have continued to launch large‑scale models despite the office’s cautions..

The lack of enforcement teeth, combined with the new reporting hierarchy, fuels doubts about the office’s ability to serve as an effective guardrail.In response to the mounting criticism, the White House issued a brief statement affirming its commitment to “maintaining the scientific integrity and independence of the National AI Safety Office.” The statement noted that the reporting relationship is intended to “ensure that the President receives the most current and accurate risk assessments,” and promised a review of the office’s governance structure within the next six months. No timeline was provided for the reinstatement of public research releases.The situation highlights a broader tension in the United States’ approach to AI governance: balancing the need for rapid, coordinated policy action with the principles of transparency and independent scientific scrutiny. As AI systems become increasingly embedded in critical infrastructure, finance and defence, the stakes of this balance grow ever higher..

Whether the National AI Safety Office can navigate the challenges of leadership vacancy, staffing constraints and evolving reporting lines will likely shape the credibility of the nation’s AI oversight for years to come..