GPT-56T 861 —
MUSE-SPK 837 —
GPT-56SC 789 -0.1%
GLM-5 781 —
CL-OP55X 779 -0.1%
GROK-46H 779 -0.1%
QWEN-38X 748 —
GPT-6A 743 —
KIMI-K3X 742 —
CL-FAB5H 697 -0.1%
CL-OP5H 674 -0.1%
GEM-38FH 672 —
CL-OP5X 669 -0.1%
CL-OP55H 667 -0.1%
CL-OP46H 656 -0.2%
CL-OP47H 647 -0.2%
GPT-56S 617 -0.2%
GEM-37FH 609 -0.2%
GEM-36FH 592 -0.2%
CL-OP48H 587 -0.2%
CL-OP47 580 -0.2%
GEM-35FH 579 -0.2%
GPT-55H 540 -0.2%
INKL 531 —
GEM-31P 511 -0.2%
CL-OP46 498 —
GEM-3P 498 —
CL-OP48 492 —
GPT-52 464 —
GPT-55 423 —
GPT-56T 861 —
MUSE-SPK 837 —
GPT-56SC 789 -0.1%
GLM-5 781 —
CL-OP55X 779 -0.1%
GROK-46H 779 -0.1%
QWEN-38X 748 —
GPT-6A 743 —
KIMI-K3X 742 —
CL-FAB5H 697 -0.1%
CL-OP5H 674 -0.1%
GEM-38FH 672 —
CL-OP5X 669 -0.1%
CL-OP55H 667 -0.1%
CL-OP46H 656 -0.2%
CL-OP47H 647 -0.2%
GPT-56S 617 -0.2%
GEM-37FH 609 -0.2%
GEM-36FH 592 -0.2%
CL-OP48H 587 -0.2%
CL-OP47 580 -0.2%
GEM-35FH 579 -0.2%
GPT-55H 540 -0.2%
INKL 531 —
GEM-31P 511 -0.2%
CL-OP46 498 —
GEM-3P 498 —
CL-OP48 492 —
GPT-52 464 —
GPT-55 423 —
← Back to feed

US Federal Agencies Are Using AI to Write Regulations — DOT Names Google Gemini Its Rulemaking Tool

The Trump administration has moved AI from an aspirational tool into the active federal regulatory process. The Department of Transportation has reportedly been using Google Gemini to write proposed rules — language it described internally as the “point of the spear” for federal AI adoption. Multiple agencies have since disclosed AI use cases that touch the rulemaking chain directly: sorting public comments on proposed rules, drafting regulatory documents, and identifying regulations for rescission.

The Department of the Treasury has reported an AI tool that both flags regulations for elimination and drafts replacement language. DOGE, earlier in the administration, proposed using a large language model to rescind approximately half of all federal regulations in a single pass — a proposal that was piloted in limited form before drawing legal challenges.

The Administrative Logic

The Trump administration is navigating a structural contradiction. It is simultaneously cutting the federal workforce through DOGE-driven reductions and eliminating entire agencies, while pursuing ambitious regulatory goals: rescinding numerous existing rules and asserting federal power in new domains, including AI preemption of state law. Doing more with fewer federal employees requires automation. AI is the chosen instrument.

Agencies are self-reporting AI use cases under an Office of Management and Budget directive. The publicly disclosed list now includes:

  • DOT: Using Gemini for drafting proposed rules and preambles
  • Treasury: AI-assisted regulation identification and document drafting
  • Multiple agencies: Comment classification tools to sort and tag public submissions on proposed rules
  • DOGE: LLM-based regulatory inventory scanning for rescission candidates

The OMB directive frames these as “efficiency tools” under existing agency authority — a legal theory that has not yet been tested in court.

The Administrative Procedure Act requires that proposed rules reflect the considered judgment of agency officials and provide a rational basis on the record. Whether an AI-generated draft satisfies the “reasoned explanation” standard under Motor Vehicle Manufacturers v. State Farm — the controlling precedent for arbitrary-and-capricious review — is an open question. Courts have not ruled on AI-drafted rulemaking, and no circuit split yet exists.

The hallucination risk is a specific concern. LLMs generate plausible-sounding regulatory text that can contain incorrect citations to existing statutes, misstatement of agency authority, or internally inconsistent provisions. Treasury’s AI-drafted documents would undergo human review before publication, but agencies have not disclosed what that review process entails or whether it is sufficient to catch model errors before documents enter the Federal Register.

Comment classification raises a separate set of issues. The APA requires agencies to consider and respond to significant comments on proposed rules. AI tools that sort or summarise public comments — without surfacing every substantive point — could create procedural vulnerabilities: if a significant comment is classified incorrectly and not addressed in a final rule, challengers would have grounds to seek vacatur.

The Google Gemini Question

DOT’s choice of Gemini is commercially notable. Google has been competing aggressively for federal AI contracts against Microsoft/OpenAI’s established GSA and FedRAMP position. A named DOT deployment — called out publicly by the agency rather than disclosed through procurement records — represents a marketing win and a foothold in a procurement category that tends to expand once established.

Gemini 3 Pro and Gemini 2.5 Pro are both available under Google’s federal terms. It is not clear from available disclosures which version DOT is using or whether it is operating under a negotiated data-handling agreement.

What Comes Next

Litigation is the most likely near-term forcing function. A regulated industry participant or public interest organisation with standing to challenge a specific rule drafted with AI assistance could test the APA question in federal court within months of a final rule’s publication. The outcome would either validate AI-assisted rulemaking as legally sound or create a precedent requiring explicit agency disclosure and heightened review standards.

Separately, Congress has not enacted any rules governing AI use in federal rulemaking. The White House’s National AI Legislative Framework, released last month, addressed federal preemption of state AI laws but was silent on intra-government AI use. That gap means agencies are operating on executive guidance alone — guidance that changes with each administration.