DOGE uses AI tool SweetREX to slash needless government regulation

Efforts to gut regulation across the US government using Artificial Intelligence (AI) are well underway. On Wednesday, the Office of the Chief Information Officer at the Office of Management and Budget hosted a video call to discuss an AI tool being used to cut federal regulations, which the office called SweetREX Deregulation AI. The tool, which is still being developed, is built to identify sections of regulations that aren’t required by statute, then expedite the process for adopting updated regulations.

The development and rollout of what is being formally called the SweetREX Deregulation AI Plan Builder, or SweetREX DAIP, is meant to help achieve the goals laid out in President Donald Trump’s “Unleashing Prosperity Through Deregulation” executive order, which aims to “promote prudent financial management and alleviate unnecessary regulatory burdens.”

Industrial-scale deregulation is a core aim laid out in Project 2025, the document that has served as a playbook for the second Trump administration. The so-called Department of Government Efficiency (DOGE) has also estimated that “50 percent of all federal regulations can be eliminated,” according to a July 1st 2025, PowerPoint presentation obtained by The Washington Post.

To this end, SweetREX was developed by associates of DOGE operating out of the Department of Housing and Urban Development (HUD). The plan is to roll it out to other US agencies. Members of the call included staffers from across the government, including the Environmental Protection Agency, the Department of State, and the Federal Deposit Insurance Corporation (FDIC), among others.

Christopher Sweet, a DOGE affiliate who was initially introduced to colleagues as a “special assistant” and who was until recently a third-year student at the University of Chicago, co-led the call and was identified as the primary developer of SweetREX (thus, its name). He told colleagues that tools from Anthropic and OpenAI will be increasingly utilized by federal workers and that “a lot of the productivity boosts will come from the tools that are built around these platforms.”

Sweet said that for SweetREX, they are “primarily using the Google family of models, so primarily Gemini.”

A spreadsheet detailed how many words could be eliminated from individual regulations and gave a percentage figure indicating how noncompliant the regulations were; how that percentage was calculated was unclear. At the time, Sweet did not respond to a request for comment, and a HUD spokesperson said the agency does not comment on individual personnel.

Leading Wednesday's call alongside Sweet was Scott Langmack, a DOGE-affiliated senior adviser at HUD and, according to his LinkedIn profile, the COO of technology company Kukun - a proptech firm that is, according to its website, “on a long-term mission to aggregate the hardest to find data.”

While Sweet led the development side of SweetREX, Langmack said he was taking point on demoing the tool for different agencies and pitching them on its benefits. He claimed, for example, that the tool is capable of reducing the time spent reviewing and proposing edited regulations from months to just a few hours or days.

The “decision tool” feature within SweetREX flags sections of regulations that it deems extraneous according to relevant statutes, according to Sweet and Langmack. US government attorneys and policymakers can then review the AI platform’s proposed changes and make adjustments, Sweet and Langmack said. The tool will also create a draft of the AI-altered regulation for review.

Before new or updated regulations are adopted, an agency may publish an “Advance Notice of Proposed Rulemaking,” which allows any member of the public to comment on the changes to the regulation. In addition to flagging clauses in regulation as extraneous, SweetREX will also have the ability to group the public’s comments on proposed changes to federal rules into AI-determined “buckets,” according to details shared on the call. The tool will further categorize the types of people or entities submitting comments, such as whether they’re an individual or a “sophisticated” corporate commenter.

Last month, a HUD worker said that combing through the regulations flagged by the AI to explain why they might still be needed was “painful.” However, several people who asked questions during Wednesday's call praised SweetREX as “awesome” and “great.”

During the call, a person in the video meeting identified as Steve Davis piped in.

“Would it be possible to open source software and put it up on GitHub?” he asked. (The answer was, essentially, “maybe.”) Steve Davis is also the name of Musk’s top lieutenant, who earlier this summer left his post as a special government employee involved with running DOGE and returned to the private sector as the president of Musk’s Boring Company - a title he did not abdicate during his time in government. After his departure, DOGE staffers were reportedly made uncomfortable by him seemingly continuing to run DOGE despite not working for the government.