Illustration: the difference between human and artificial intelligence
Canva Design Studio
An apparent Trump administration plan to dramatically reduce the amount of federal regulation—reported in a recent Washington Post article and outlined in a leaked July 2025 Department of Government Efficiency (DOGE) presentation—is bold, sweeping, and a bit naïve. While the proposal rightly identifies a problem with excessive and ineffective regulations and the potential of artificial intelligence (AI) to identify necessary reforms, it does not seem to appreciate the legal and procedural steps involved in deregulation.
As someone who has long advocated for greater regulatory restraint, I believe automated tools can play a valuable role in identifying rules that may not be required by statute, impose unnecessary burdens, or hinder innovation and growth. Used appropriately, AI can help agencies screen for outdated, burdensome or ineffective regulations and flag areas for deeper analysis. But automation is not a substitute for the deliberative, transparent, and legally-grounded processes required by the Administrative Procedure Act and longstanding executive orders. Ignoring those requirements may make deregulatory actions short-lived.
DOGE expects the AI tool will cut the time to deregulate by 93%
The leaked DOGE slide deck estimates that the “DOGE AI Deregulation Decision Tool” will analyze more than 200,000 federal regulations and eliminate 50% that are “not statutorily required.” (It’s not clear how DOGE is defining regulation—perhaps as sections of the Code of Federal Regulations.) The deck estimates that, without the AI tool, each regulatory section would require 36 person-hours to 1) determine whether it is required by statute (2 hours), 2) draft and propose a revised regulation (10.5 hours), 3) analyze and respond to 100,000 public comments (12.5 hours), and 4) draft and publish a final rule (11 hours). It says the AI tool would cut that time by 93%.
I’m first struck by how unrealistic the estimate of the human time involved in these steps is. Regulatory justifications are detailed documents, often hundreds or thousands of pages long, that lay out not only the legal basis for an action, but a factual, scientific, and economic record in support of the chosen approach. This usually takes years, not hours or days. Once finalized, affected parties can sue; and for a deregulatory action, courts will compare the justifications for the original rule with those of the rescission. To withstand judicial scrutiny, the docket supporting the deregulatory action must be more robust and persuasive than the original, which would certainly involve more than 36 hours to develop. In light of the Loper Bright Supreme Court decision, there may be some persuasive legal arguments that an earlier statutory interpretation was incorrect, but it seems unlikely that such arguments will prevail for anywhere near 50% of all the rules currently on the books.
On the one hand, underestimating the human time to revise regulations may mean the 93% estimate understates the savings AI can deliver. On the other, it is unlikely that AI tools alone can do the drafting and reasoning needed to create a defensible regulatory docket. A White House official told the Post that “the DOGE experts creating these plans are the best and brightest in the business,” and I have no reason to doubt that. But, as Elon Musk discovered to his frustration, the business of government isn’t the same as the business of business; and a move-fast-and-break-things approach is unlikely to yield lasting achievements. The DOGE slide deck suggests that agencies will use the AI tool to complete their deregulation lists by September 1, 2025 and that DOGE will take it from there to “roll-up a delete list” and automate the steps—from drafting Federal Register notices to evaluating public comment and publishing a final rule—to delete 50% of all federal regulations by January 20, 2026. It does not mention the litigation that is bound to ensue once the deregulatory actions are final.
Where AI can help
Reducing unnecessary regulations that hinder innovation, opportunity and economic growth is an admirable goal that new AI tools may be uniquely qualified to support. In particular, AI can efficiently screen the almost-200,000 pages of regulatory code to identify inefficiencies and redundancies. And, as the Commonwealth of Virginia has demonstrated, it can highlight regulatory requirements that go beyond statutory mandates. AI is also proving valuable in reviewing and categorizing public comment.
Another possibility apparently not under DOGE consideration would be to harness AI to support evaluation of whether regulations are achieving intended results. Agencies have generally failed to measure actual regulatory outcomes against expectations, but new generative and agentic models, as well as more traditional AI data analytics tools, may be able to support such retrospective review. Armed with that knowledge, agencies would be in a strong position to justify rescinding, modifying, or retaining regulatory frameworks.
Agencies must lay the groundwork to achieve lasting success
Using AI, this administration has an extraordinary opportunity to weed out regulations that are not improving social welfare. To succeed, however, it must provide reasoned justifications for rescissions—including by conducting regulatory impact analysis sufficient to convince the courts of the merits of any changes. It must also take public notice and comment seriously.
In designing a workable process, DOGE should ask whether its approach would work if a subsequent administration used it to impose new regulations. If an agency developed a “generative rule-writing model,” we can imagine that many procedural safeguards would be required before the public could be asked to obey the resulting rules. For better or for worse, deregulatory actions need similar procedural safeguards.
Without thoughtful analysis and serious evaluation of public comment—even if DOGE does succeed in finalizing actions to remove large swaths of regulation—it may have won a Pyrrhic victory. For their deregulatory actions to stick, agencies will need to do the hard work of developing a record that withstands judicial scrutiny. AI can definitely support that effort, but detailed analysis and human judgment are essential. Regulatory reform is overdue, but to achieve lasting reforms, we must invest in tools that support—not override—the rule of law.
