The Trump administration built an AI-powered tool specifically designed to hunt down and eliminate federal regulations, according to newly released government documents obtained through a Freedom of Information Act request by legal advocacy group Democracy Forward.

The tool, called SweetREX, was developed by a self-described Elon Musk acolyte and pitched to government employees as a system for "regulation extermination." The documents, reviewed exclusively by The Lever, reveal for the first time how the AI program was trained and what criteria it uses to target specific rules for elimination.

What SweetREX Was Built to Do

According to the records, SweetREX was programmed to identify and flag regulations that imposed costs on private enterprises, limited business innovation, or used race-based classifications. Using those parameters, the system could process more than 100,000 public comments in less than half an hour, a task that would take human reviewers weeks or months.

But SweetREX was not just a search engine for inconvenient rules. The documents show the tool was designed to go further: it could also draft new regulatory language and propose replacement statutes aligned with the administration's deregulation agenda. In other words, the AI was not merely identifying regulations to cut. It was writing new ones.

The Shortcuts That Should Concern Everyone

"The documents reveal, for the first time, the shortcuts this AI tool takes when deciding whether a regulation is legally required and whether its burdens outweigh benefits to the public," said Daniel McGrath, senior oversight counsel at Democracy Forward.

That framing matters. Federal regulations often exist because Congress mandated them. Environmental protections, workplace safety standards, financial disclosure requirements: many of these rules are not optional policy choices but legal obligations. An AI tool optimized to flag regulations that "impose costs on private enterprises" does not distinguish between a regulation that is inconvenient and one that is legally required. It treats both the same way.

The DOGE Connection

The tool is linked to Musk's broader influence on the Trump administration's deregulatory push. The Department of Government Efficiency, or DOGE, has been at the center of sweeping federal workforce reductions, agency restructurings, and budget cuts. SweetREX represents the technological arm of that campaign: using AI to do at scale what human policy analysts could not do fast enough.

The speed is the point. Processing 100,000 public comments in 30 minutes means the tool could theoretically review and summarize the entire public comment period for a major regulation in the time it takes to eat lunch. The question is whether "summarize" means faithfully represent public input or filter it through pre-set ideological criteria. The FOIA documents suggest the latter.

AI as Governance Infrastructure

SweetREX is not the first time AI has been proposed as a tool for government efficiency. But it may be the first documented case where an AI system was explicitly designed to write replacement legislation. That crosses a line from automation into governance. When an AI tool trained on deregulatory criteria starts proposing new statutory language, the question is no longer whether AI can help government work faster. The question is who programmed it and what they told it to optimize for.

The Trump administration has already used AI to evaluate federal workers for termination, according to prior reporting. DOGE teams deployed AI tools to assess agency budgets and identify "non-essential" spending. SweetREX extends that logic to the regulatory code itself: feed in the entire body of federal rules, apply cost-to-business criteria, and generate a kill list.

What Happens Next

Democracy Forward obtained these documents through FOIA, which means they represent only what the government was willing to release. The full scope of SweetREX's deployment, how many regulations it flagged, whether its recommendations were acted on, and who had final approval authority, remains unknown.

What is known: the tool exists. It was built to identify regulations that help the public but cost businesses money. It was designed to propose replacements. And it was pitched to government employees by someone aligned with the same deregulatory movement that has already gutted federal agency staffing.

When AI was used to screen resumes, we debated bias. When AI was used to moderate content, we debated censorship. Now AI is being used to write the rules that govern 330 million Americans. The debate has not caught up.

First reported by The Lever, based on documents obtained by Democracy Forward through a Freedom of Information Act request.