DOGE AI Aims to Cut Half of Federal Regulations

The Department of Government Efficiency (DOGE) is developing an artificial‑intelligence system intended to identify and recommend the repeal of roughly half of the federal regulations in the United States [1]. Internal documents describe a “DOGE AI Deregulation Decision Tool” that has already flagged about 100,000 rule sections as unnecessary, with the agency projecting large compliance savings and new revenue if those rules are removed [1].

Key takeaways

• DOGE’s AI tool has flagged ~100,000 federal rules for possible repeal, representing about 50 % of the Code of Federal Regulations [1].
• The agency estimates $1.5 trillion in annual compliance savings and up to $1.1 trillion in new government revenue from deregulation [1].
• A pilot at HUD, called SweetREX, would have the AI suggest “keep, delete, or partial delete” each rule, with attorneys providing final approval [2].
• Critics warn the AI’s default deregulatory bias and unrealistic time‑savings assumptions could undermine statutory review and congressional intent [2].
• Parallel efforts to use AI in workforce reductions, such as the modified AutoRIF system, raise additional legal and employee‑rights concerns [3].

AI‑Driven Deregulation Pilot at HUD

In the summer of 2025 DOGE presented a PowerPoint to the Department of Housing and Urban Development (HUD) describing SweetREX, a tool named for DOGE associate Christopher Sweet [2]. The presentation outlined a multistep workflow: the AI would analyze every HUD regulation, assign a recommendation to keep, delete, or partially delete the rule, and then hand the output to agency attorneys for review. DOGE emphasized that the AI would “automate all the most time‑consuming steps in deregulation” while leaving “program groups” in control of final decisions [2]. However, a source familiar with the FOIA documents noted that the instructions explicitly set a “default assumption” that rules should be rescinded, raising concerns that the AI could be steered toward a deregulatory outcome regardless of merit [2].

The same source questioned the feasibility of the tool’s claimed efficiency gains. DOGE’s PowerPoint suggested that SweetREX could review 100,000 public comments in 2.4 hours, a stark contrast to the agency’s current estimate of 36 hours of human review per rule [2]. Former Department of Labor staffers argued that such timelines ignore the wide variation in comment complexity and often require weeks of analysis, making the AI’s projected speed unrealistic [2]. Additionally, the documents reference the Supreme Court’s 2024 Loper Bright decision as a benchmark for “determining compliance,” yet experts say the ruling offers only vague guidance for agency authority, which may be beyond the capacity of a language model to interpret [2].

AI and Federal Workforce Decisions

While the deregulation initiative draws most attention, DOGE is also experimenting with AI in personnel management. Reports indicate that the agency has been modifying a legacy program called AutoRIF, originally designed to assist human decision‑makers in Reductions in Force (RIF) events [3]. The Trump administration reportedly issued a memorandum ordering agencies to submit RIF plans, and an internal DOGE email threatened termination for workers who failed to account for their time, with responses intended to be fed into a large language model to decide employment status [3]. These actions have sparked debate over whether AI‑driven termination tools comply with federal employment protections, including the “for‑cause” standard that requires legitimate reasons for dismissal [3].