Experts warned Monday that the Trump administration's use of artificial intelligence to accelerate deregulation could expose federal agencies to greater legal risk, as AI-driven decisions become harder to explain and defend in court.
At a panel sponsored by the Coalition for Sensible Safeguards, legal scholars, economists, and technology advocates raised concerns that reliance on AI tools - from analyzing public comments to drafting regulatory language - may undermine the rational basis required for agency actions. The discussion highlighted how advanced AI systems, often described as black boxes, can obscure the reasoning behind a rule, making it vulnerable to judicial challenge.
AI already embedded in agency work
Cary Coglianese, a regulation professor at the University of Pennsylvania, said federal agencies have been using AI for years to sort through the thousands of public comments proposed rules receive. "Back in the day, contractors would literally photocopy comments, clip them out and sort them in different categories - and do what essentially natural language processing tools can do now so quickly and rather effectively," he said.
But as the technology advances, panelists warned that AI could take over more steps in the rulemaking process, potentially supercharging efforts to slash regulations and downsize the federal workforce. Ben Zipperer, a senior economist at the Economic Policy Institute, said there is no market mechanism to correct poor AI implementation in government. "A private sector company could replace all their workers with AI, but then their product quality would probably suffer, and they'll lose sales. But there's no profitability check on the public sector," he said. "So if the public sector uses AI as a way or an excuse to make public servants' jobs worse or to cut public services, there's not a non-political check on that, and public services just suffer."
Risks of biased outputs and ignored feedback
The panel pointed to a high-profile example: Department of Government Efficiency staffers used ChatGPT to assess National Endowment for the Humanities grants and flag those related to diversity, equity and inclusion, which were targeted for elimination. Jonathan Walter, a senior policy counsel at the Leadership Conference on Civil and Human Rights, said officials could similarly use AI to disregard public comments they oppose. "It would have been just as easy for them to say, 'Hey, we're just going to delete or ignore any comment with any mention of race or disparate impact or gender or whatever protected class might have a stake in a rulemaking,'" he said.
J.B. Branch, a technology expert at Public Citizen, raised concerns about the specific AI systems federal employees might use, singling out Grok, which has been criticized for generating racist comments as well as responses aligned with the views of its creator, Elon Musk. "If someone in the federal government is using Grok, there is a deep concern that whatever output it's providing could be a potentially very biased output," he said.
Last year, the president signed an executive order directing agencies to procure "ideologically neutral" and "truth-seeking" AI that is not "woke." But the administration's aggressive deregulatory push - which officials have said will "promote liberty, unleash American energy dominance, preserve products consumers love and eradicate the ideology of [DEI]" - raises questions about whether such standards can be enforced when the technology itself is opaque.
Judicial review and the black box problem
Lauren McFerran, executive director of the AFL-CIO Tech Institute, said an overreliance on AI during rulemaking could backfire by inviting more judicial scrutiny. Courts require agencies to show a rational connection between the evidence, their reasoning, and the final decision. With advanced AI systems that learn from their own processes, that connection can become invisible. "That creates - the term that tends to be used is a black box. You don't really necessarily know what steps the product undertook to get to certain outcomes," she said. "I think that could be potentially problematic for a court that's trying to review whether the agency, in fact, engaged in the reasonable decision-making process that it was supposed to."
That risk extends beyond the current administration's deregulatory goals. Any agency that uses AI to support rulemaking may face challenges if the underlying logic cannot be explained. Legal experts say the Administrative Procedure Act demands a clear explanatory record, and AI's opacity could undermine that requirement.
Why this matters for government employees
For federal workers and managers involved in rulemaking, the message is clear: AI tools can speed up comment analysis and drafting, but their use must be carefully documented and justifiable. Courts will look for a transparent, reasoned decision-making process, not an automated output that no one can fully explain. As agencies integrate AI into regulatory work, gaining practical skills through AI for Policy Makers training can help staff understand both the technology's capabilities and its legal risks. Failing to build that expertise could lead to rules being overturned - and the very deregulatory wins the administration seeks being lost in court.
Your membership also unlocks:
