DOGE Built An AI To Delete Half Of Federal Regulations. Will It Work?

The Washington Post recently revealed that the Department of Government Efficiency, known by its acronym DOGE, is developing and deploying an artificial intelligence tool designed to eliminate as much as half of all federal regulations. According to internal government documents obtained by the Post, the “DOGE AI Deregulation Decision Tool” has already flagged approximately 100,000 federal rules that are not required by law. The hope is that AI can also be used to automate the most labor-intensive parts of the regulatory repeal process, so that a major deregulatory effort will be underway by the first anniversary of President Trump’s second term in office.

The internal documents reviewed by the Post include a PowerPoint presentation dated July 1, 2025, which lays out the assumptions and expectations driving the initiative. That document proposes that roughly 50 percent of the Code of Federal Regulations, or around 100,000 individual rule sections, could be repealed without violating any statutory obligations.

DOGE estimates that the total cost of regulatory compliance in the United States is around $3.1 trillion annually. Stripping out rules that are not legally required could generate up to $1.5 trillion in annual compliance savings. Additionally, deregulation could unlock $600 billion in new investment and $1.1 trillion in new government revenue from untapped economic activity, according to DOGE. These are headline-grabbing numbers, though the methodology behind them isn’t fully clear from the presentation.

The AI tool at the center of DOGE’s efforts is built to scan large volumes of regulatory text, compare those rules against their enabling statutes, and determine whether each section is mandatory or discretionary. In theory, if a rule goes beyond what Congress has required, or simply rephrases statutory language without adding any interpretive value, it could be marked for deletion. In practice, the task could get more complicated, as legal interpretation often lies in gray areas that AI algorithms may struggle to understand.

DOGE’s analysis suggests that repealing 100,000 regulations through traditional means would require about 3.6 million man-hours of legal and policy work. This includes the time required to research the law, draft proposed repeal notices, review and respond to public comments, and finalize repeal orders. Using the AI tool, the presentation claims this workload could be reduced by 93 percent. Most of the reduction would come from automation of tasks such as generating initial legal drafts and analyzing public comment responses.

The AI tool has already been deployed at a couple of agencies. At the Department of Housing and Urban Development, it was used to evaluate over 1,000 regulatory sections in under two weeks. A similar effort at the Consumer Financial Protection Bureau reportedly relied on the AI to draft 100 percent of the agency’s deregulation proposals. Whether these pilot programs can be replicated across the government remains to be seen. Some agency officials have voiced concerns about the accuracy of the tool’s legal interpretations. According to the Post, HUD staffers found that the AI misread the law in a number of cases.

In the short term, the process to identify and repeal rules will inevitably require human oversight. Regulations exist within a complex and dynamic legal environment, and understanding their interactions with statutes, case law, and enforcement practices is not something that can yet be fully delegated to machines. However, DOGE has thus far framed the tool as an assistant that can reduce labor burdens and help focus legal expertise where it’s most needed, not as a substitute for human judgment.

Even so, the legal terrain surrounding this initiative is uncertain. The Administrative Procedure Act sets out specific requirements for repealing federal rules, including public notice and comment, reasoned explanation from the agency, and a prohibition on actions deemed “arbitrary and capricious.” Courts have been skeptical in the past of deregulatory efforts that fail to meet these standards, and the use of AI could trigger new legal questions about what constitutes adequate administrative reasoning.

There is also the matter of institutional resistance. The Post reports that some career staff have expressed reluctance to adopt DOGE’s approach, citing concerns about outsourcing regulatory judgments to machines. Workforce reductions in the federal government may also have left agencies understaffed to manage a sudden surge in AI-generated repeal proposals.

Still, the plan marches forward. DOGE intends to train all federal agencies on the use of its tool by the end of July. Each agency is expected to submit a finalized list of regulatory sections targeted for elimination by September 1. The timeline culminates in January 2026, when agencies are scheduled to submit repeal packages for review and approval. The slide deck calls this the “Relaunch America” initiative.

Whether that reboot is successful, or even legally viable, is an open question. It’s possible that the courts will reject major parts of the initiative if AI-generated repeal notices fail to meet administrative law standards. But some degree of litigation may be inevitable, and perhaps even useful. The judicial system could play a constructive role in clarifying the extent to which AI can be used in regulatory decision-making. As with past episodes of administrative innovation, it often takes a few lawsuits to define the rules of the game.

In the end, this DOGE initiative is a test of the federal government’s appetite for experimentation. It is a wager that some portion of the administrative state is ripe for digitization. But it may also reveal the limits of current institutional capacity. In the coming months, we will find out which of limits are malleable, and which are not.

The era of algorithmic governance may not be fully upon us, but it is no longer a thought experiment confined to white papers. It has arrived in the form of actual updates to government code. Whatever happens, the experiment is worth watching.