Source: FedScoop

DOGE presented SweetREX, an AI-driven tool intended to analyze all Department of Housing and Urban Development regulations and recommend whether to “keep, delete, or partial delete” each rule, based on a PowerPoint delivered at HUD last summer. Documents obtained via Freedom of Information Act requests by Democracy Forward and shared with FedScoop, with earlier Wired reporting referenced, lay out a multistep process and explicit instructions that frame deregulation as the default case, instructing staff to treat proposed regulatory eliminations as the starting position and to note disagreements and rationales in a provided spreadsheet.

The tool’s outputs were described as subject to attorney review and final decisions by program staff, while DOGE’s pitch emphasized automating time-consuming deregulatory steps and freeing HUD legal and policy teams for “high-value work.” The FOIA materials do not make clear whether any HUD rules were actually written or rescinded based on SweetREX recommendations, and sources quoted in the documents expressed concerns that large language models lack the capability to perform nuanced legal analyses and that prompting could bias results toward rescission. DOGE’s presentation claimed SweetREX could cut rulemaking time by at least 90% and dramatically speed review of large volumes of public comments, but former agency sources questioned those baseline assumptions and whether an LLM can determine compliance with complex legal standards cited in the documents; HUD’s press office did not respond to questions in the extracted text.