US AI safety chief resigns amid mounting policy battles

Chris Fall resigned as director of the Commerce Department’s Center for AI Standards and Innovation after three months in the job, and Arvind Raman has been named acting chief. The change lands as Washington is still trying to move from warnings about AI risk to actual oversight that can be enforced.

CAISI sits at the center of that effort. The office works with academics and developers to identify and mitigate risks from cutting-edge AI systems, including hallucinations, misinformation, cybersecurity vulnerabilities, discrimination and misuse in biology, surveillance and autonomous decision-making. A leadership vacancy can slow the standards work, internal reviews and interagency coordination that give the government’s AI safety push its practical shape.

The departure matters because the agency’s role is not ceremonial. It is one of the federal government’s main technical arms for understanding how frontier models behave and where they can fail, and its work can influence conversations inside Commerce, Homeland Security, Defense and the White House. With AI tools moving into consumer software, office systems, health care and defense applications, the government’s ability to keep pace depends in part on whether this office has stable leadership.

The timing also reflects a broader shift in U.S. policy. The Trump administration has moved in a different direction on AI security, and earlier this year Washington was already showing signs of a broader safety pivot. That leaves companies in a familiar tension: many want less uncertainty and a more predictable regulatory path, while critics say any easing of pressure risks widening the gap between rapid technical development and public safeguards.

Fall’s exit is not the first turnover at the office. A February 2025 report on the U.S. AI Safety Institute described a previous director leaving the role, a sign that the leadership bench around America’s AI safety apparatus has remained unsettled. The latest resignation could deepen concerns that the government is still struggling to build durable oversight just as frontier systems are becoming more capable and more widely deployed.

Commerce said a new director will be announced in the coming weeks. Until then, the office’s direction rests with an acting chief at a moment when the politics of AI safety are tightening, not easing.