When ‘Good Enough’ Isn’t Safe: AI and Federal Rulemaking
Federal employees are increasingly being asked to do more with less—fewer staff, tighter deadlines, and higher stakes. A recent report that the Department of Transportation plans to use Google’s Gemini AI to help draft federal transportation regulations adds a new and troubling layer to that reality. These are not internal memos or briefing notes. These are safety rules that determine how airplanes fly, pipelines operate, and trains carry hazardous materials.
The concern is not technology itself. Many agencies already use advanced tools responsibly. The concern is the philosophy reportedly driving this shift: leadership signaling that rules do not need to be “perfect,” only “good enough,” and that speed and volume of rulemaking matter more than precision. In safety regulation, that mindset carries real risk.
Why Rulemaking Is Different From Other Writing
In federal rulemaking, the writing is the substance. Regulatory text reflects statutory interpretation, scientific judgment, and policy reasoning. Courts do not treat it as filler. When injuries occur, judges examine the words closely and ask whether the agency acted lawfully, reasonably, and based on evidence in the administrative record.
Large language models are designed to generate fluent text, not to ensure legal accuracy. They can misstate statutes, invent citations, gloss over scientific uncertainty, or subtly alter definitions in ways that appear polished but are wrong. When human reviewers are rushed—or when staffing levels are already thin—those errors can slip through.
The Compounding Risk of Institutional Loss
This issue does not exist in a vacuum. Across government, agencies are operating with fewer subject-matter experts, fewer experienced attorneys, and fewer reviewers with decades of institutional knowledge. Those individuals historically served as guardrails against catastrophic regulatory mistakes.
If the process becomes “AI drafts, humans proofread,” agencies are no longer governing—they are editing. That distinction matters. One faulty assumption in an aviation rule or one vague definition in a pipeline standard is not a technical error. It can translate into injuries, environmental damage, or loss of life.
Legal Vulnerability Becomes a Safety Problem
Weak AI-assisted rulemaking also creates legal exposure. If regulations are challenged and courts find the reasoning arbitrary or inadequately supported, those rules can be vacated. When safety regulations are struck down, the public does not become safer—it becomes less protected.
For federal employees, this creates a familiar and unfair dynamic: pressure to move fast now, followed by blame later when something goes wrong.
Practical Takeaways for Federal Employees
For rule writers and reviewers, speed should never substitute for accuracy. Concerns about AI-generated content should be documented in writing. Drafts, instructions, and decision-making chains should be preserved.
For supervisors, HR, ER/LR professionals, unions, and oversight staff, this is a workplace integrity issue. Employees should not be disciplined or marginalized for insisting on rigor in life-and-death regulatory work.
And for anyone participating in or monitoring public comment periods, those comments matter. They build the administrative record that slows down bad ideas and strengthens defensible rules.
Legal Disclaimer: The information provided in this article is for informational purposes only and should not be construed as legal advice. While I am a federal employment attorney, this post does not create an attorney-client relationship. Every situation is unique, and legal outcomes depend on specific facts and circumstances.