Forgot your password?
typodupeerror
AI United States

DOT Plans To Use Google Gemini AI To Write Regulations (propublica.org) 62

The Trump administration is planning to use AI to write federal transportation regulations, ProPublica reported on Monday, citing the U.S. Department of Transportation records and interviews with six agency staffers. From the report: The plan was presented to DOT staff last month at a demonstration of AI's "potential to revolutionize the way we draft rulemakings," agency attorney Daniel Cohen wrote to colleagues. The demonstration, Cohen wrote, would showcase "exciting new AI tools available to DOT rule writers to help us do our job better and faster."

Discussion of the plan continued among agency leadership last week, according to meeting notes reviewed by ProPublica. Gregory Zerzan, the agency's general counsel, said at that meeting that President Donald Trump is "very excited about this initiative." Zerzan seemed to suggest that the DOT was at the vanguard of a broader federal effort, calling the department the "point of the spear" and "the first agency that is fully enabled to use AI to draft rules." Zerzan appeared interested mainly in the quantity of regulations that AI could produce, not their quality. "We don't need the perfect rule on XYZ. We don't even need a very good rule on XYZ," he said, according to the meeting notes. "We want good enough." Zerzan added, "We're flooding the zone."

These developments have alarmed some at DOT. The agency's rules touch virtually every facet of transportation safety, including regulations that keep airplanes in the sky, prevent gas pipelines from exploding and stop freight trains carrying toxic chemicals from skidding off the rails. Why, some staffers wondered, would the federal government outsource the writing of such critical standards to a nascent technology notorious for making mistakes? The answer from the plan's boosters is simple: speed. Writing and revising complex federal regulations can take months, sometimes years. But, with DOT's version of Google Gemini, employees could generate a proposed rule in a matter of minutes or even seconds, two DOT staffers who attended the December demonstration remembered the presenter saying.

This discussion has been archived. No new comments can be posted.

DOT Plans To Use Google Gemini AI To Write Regulations

Comments Filter:
  • Federal Regulations (Score:5, Interesting)

    by Retired Chemist ( 5039029 ) on Monday January 26, 2026 @05:28PM (#65950910)
    Having had to read federal regulations in my job, I can state with authority that 90% or more of the text is boilerplate. A properly trained LLM would have no problem producing it. The small amount of material that is actually important would have to be part of the prompts in any case. As long as someone bothers to proofread the resulting document, I really do not see that it should be a problem.
    • by dknj ( 441802 ) on Monday January 26, 2026 @05:32PM (#65950920) Journal

      This is what an LLM would want us to believe. Youâ(TM)re not fooling anyone, bot

    • by taustin ( 171655 ) on Monday January 26, 2026 @05:40PM (#65950938) Homepage Journal

      As long as someone bothers to proofread the resulting document,

      And what are the odds of that? Seriously, lawyers are going to be getting disbarred soon because they don't do that, with repeat offenders who certainly know better. Government employees are even harder to fire.

      What, really, are the odds that any human eye will see these new regs before they are implemented?

      • What, really, are the odds that any human eye will see these new regs before they are implemented?

        You can always have another AI proofread the output. As an added bonus the DOT can then be made much leaner by firing the employees, as in cut out the middle man. A bit like an AI centipede!

        • by skam240 ( 789197 )

          A bit like an AI centipede!

          And I'm sure we'll get great results from a bunch of AI metaphorically shitting in each others mouths.

          • by taustin ( 171655 )

            If there's one thing where it's desirable to replace humans with AIs, it would be the Human Centipede.

      • What are the odds? Probably about the same as now.

    • Re: (Score:3, Informative)

      Yeah, but a human being can also generate it.

      Do you really think a human being can't shit out a "proposed rule" in three days? Fuck, I bet many proposed rules can be written in an hour. The issue isn't the time taken to create a proposed rule, it's the effort needed before to determine what kind of rule it should be, and the time afterwards where the rule is reviewed by safety experts.

      NONE of those parts are somehow possible to pass on to Spicy Autocomplete in any serious manner. You really want the propose

      • The fact of the matter is that writing the rule is the least of the work. Deciding what it will be is what takes sometimes years. A rule is proposed and there is a comment period and then the rule is revised based on the comments (or at least it is supposed to be, and if it isn't be prepared to hear from the ignored interest groups congressmen). Then the revised rule is commented on again through, however, many iterations it takes. The actual writing part is trivial.
      • Thanks for this, Spicy Autocomplete is my new favourite words.
        • by aitikin ( 909209 )

          Thanks for this, Spicy Autocomplete is my new favourite words.

          I prefer it as Spicy Autocorrupt, personally.

    • by gtall ( 79522 )

      With this alleged administration you actually believe anyone will proof read the result? If most is boilerplate, then build a stock of boilerplates (a stock probably already exists) and have a worker slap one on before writing new regulations. That is, if you haven't fired all the knowledgeable people and now must rely upon AI and all its ills.

      • You can bet the people affected by the regulation will read it. They have whole departments that do nothing else. Most of them also have lawyers and lobbyists.
    • What's funny is that you usually need an LLM to get past the boilerplate to understand what the regulation is actually doing.
    • I can state with authority that 90% or more of the text is boilerplate.

      Yep.

      A properly trained LLM would have no problem producing it.

      But an LLM can not do maths. It is inappropriate in this circumstance as it can not do logic. Regulations are like a math discipline. They must be coherent as a whole, not merely when looked at from certain perspectives. LLMs can not do coherency.

  • Well, that might just fit the Trump administration quite well.

  • by Waffle Iron ( 339739 ) on Monday January 26, 2026 @05:33PM (#65950922)

    calling the department the "point of the spear"

    Or more accurately: "the spout of the slop pail".

    • by Anonymous Coward

      calling the department the "point of the spear"

      Or more accurately: "the spout of the slop pail".

      The blown-out sphincter of the anus.

    • by whit3 ( 318913 )
      Oh, the department will be the other end, the butt... of the joke.

      Regulations have to be readable by affected persons and organizations,
      by judges/juries, and not be easily warped by lawyers.
      The additional constraint that they have to make sense to an AI,
      is... not likely to produce a quality product by those criteria.

      .
      It's unlikely that good correspondence with 'the will of the people' will result
      from this kind of intermediate step in the democratic process.
  • “in a matter of seconds.” Gee whiz, with those kind of requirement you can do anything. In just a matter of seconds I could fund socialized medicine, do nothing, stub my toes, or launch all our nukes. If the results don’t matter, what’s the difference? Why do anything at all? Because the only thing this is really about is making sure the rich can get richer the fastest as possible with as little interference from pesky laws and regulations they can manage.

    “No, no, no! How da
  • by Pseudonymous Powers ( 4097097 ) on Monday January 26, 2026 @05:37PM (#65950930)

    'Zerzan appeared interested mainly in the quantity of regulations that AI could produce, not their quality. "We don't need the perfect rule on XYZ. We don't even need a very good rule on XYZ," he said, according to the meeting notes. "We want good enough." Zerzan added, "We're flooding the zone."'

    This is self-satire. "Good enough"? Good enough for what? "Government work"? The point of that joke is that it's not really good enough!

    And "We're flooding the zone"? The end of that sentence is "with shit". That's not me editorializing. That's just a fact.

    If text requires no thought to write, if it contributes no semantic data, it shouldn't be written in the first place.

    The more I hear about this, the more I think that the "AI Revolution" is really just people not knowing how to type.

    • by Pseudonymous Powers ( 4097097 ) on Monday January 26, 2026 @05:38PM (#65950934)

      Also, when rules are being written thoughtlessly, the fact that you can write them faster is BAD. Everybody KNOWS that, right? Because then you've got tons of thoughless rules. That's bad! That's clearly bad!

      AUUAUAAHAAGH!

      • by test321 ( 8891681 ) on Monday January 26, 2026 @06:08PM (#65950998)

        I come here to concur on your argument on the thoughtless rules. This is so incredible it has to be an April 1st joke.

        The problem isn't even the AI, it is believing that "good enough" and "mere minutes" are a good properties for government work.
        The only interpretation I can figure is they hired incompetents who believe the problem of government is ability to draft, because they haven't studied anything useful nor have relevant experience at government agencies. The drafting is the easy part, of course, people who work for government in functions involving drafting learn that skill in law school, they can do it in mere minutes already. The question is not how fast one can draft, it is what to draft. Making sure this is the right change to make is what takes months.

        Also what is the point of "flooding" (even forgetting the "shit" part)? Governing isn't a competition on how many regulations one can publish in a day. I can't help but think they're trying to be evil in some way which isn't totally clear yet.

        • See, but that's the thing. To them, it absolutely IS a competition.

          A competition they'll approve any price paid by someone else to win at, because winning is what matters to them more than anything else, and they don't understand why you don't get that, because they lack the theory of mind to comprehend people who live outside their all-encompassing competition mindset.
      • by ratbag ( 65209 )

        I was going to spell it "ARRRGGGGHHH" but your way works just as well. When will the madness end? Either Trump or "AI" or preferably both.

    • by DarkOx ( 621550 )

      This should tell you all you need to know about not only Trump administrations view of law and administrative procedure but law in general.

      So very very much of it exists not because its good, needed, or even makes sense but for its own sake and to be purposefully vague so that it can be twisted to mean whatever the hell bureaucrats in charge want to mean at any given time.

      Honestly no free society should tolerate this. We should create a constitutional amendment that requires fines and penalties vary inverse

    • Wait, I thought these guys never met a regulation they didn't want to cut/remove..

      Clean air regulation? naaah let them breathe toxins

      Clean water regulation? Naaah its their fault if they drink poop

      AntiTrust rule? naah not if its one of the companies that donated a million to the turnips innaugeration or came and kissed his rings

      It goes on and on (oh and yeah I realize I did the rule of threes here and so um

      Anti Nebulon of Praxis 3 rule? naaah that was just something Captain Kirk said to make it sound scienc

    • The more I hear about this, the more I think that the "AI Revolution" is really just people not knowing how to type.

      The AI revolution is humans coming up with a program that caters to their desires without regard to Reality. AI will be used to reinforce beliefs rather than seek out Truth. Opiates (Soma?) would be a better usage of time and energy.

    • by nmb3000 ( 741169 )

      And "We're flooding the zone"? The end of that sentence is "with shit". That's not me editorializing. That's just a fact.

      I'm pretty sure that's the entire point behind the exercise (well, except some bonus mild cruelty associated with thoughtless and shitty rules that will inevitably be biased against the poor and powerless). They want to create terrible regulations so that they can point at them later and screech about the TERRIBLE OVERBEARING REGULATIONS. This gives them a lever in 3 years to push against these AI regulations in addition to any others they want to excise.

      This is just the techbro version of what Regan did

  • Well now the USA is using AI to write the laws there I think the meat bags there remember the correct way to talk. The phrase you guys need to learn is "I for one welcome our AI overlords".
  • This "full steam ahead damn the consequences" attitude scares me.

  • by subreality ( 157447 ) on Monday January 26, 2026 @05:58PM (#65950984)

    employees could generate a proposed rule in a matter of minutes or even seconds

    Okay, sure, it'll draft some text in minutes. You then have to review it in detail to see if it's actually what you intended, which takes at least an order of magnitude longer. You then have to validate if the idea you came up with in seconds or minutes is actually a good idea. Have you thought about second order effects? Have you considered alternatives? Have other people reviewed the ideas? Are you going to get buy-in from everyone else involved?

    If you're not doing those things, then you're just generating low-quality slop which wastes other people's time, or worse, gets rubber stamped and creates a real mess. Just slopping out more regulations faster is not a good goal.

    If you DO do those things, then the LLM has helped you shave some time off of a small portion of a much larger process. It's a useful tool for that, but let's be realistic about what the actual gains are.

    • If you're not doing those things, then you're just generating low-quality slop which wastes other people's time, or worse, gets rubber stamped and creates a real mess. Just slopping out more regulations faster is not a good goal.

      A good point- and this administration would never do anything like that

  • by pak9rabid ( 1011935 ) on Monday January 26, 2026 @06:09PM (#65951002)
    Why don't we just have AI run all of the federal government? At this point it can't do much worse than the dipshits currently in charge...
  • by oldgraybeard ( 2939809 ) on Monday January 26, 2026 @06:11PM (#65951006)
    Not do it the profitable way and let the lobbyist do it?
  • No pun intended, but maybe this will allow you to get modern automotive tech like matrix headlights not so many years after everyone else.
    • They're legal here, now, since 2021.
      The legal reasoning for why they could not be made legal earlier is included in the NHTSA rule change notice, and it will definitely make your head spin.
  • Might as well use a psudo-random-number generator to come up with randomly written laws.
  • by Above ( 100351 ) on Monday January 26, 2026 @08:02PM (#65951222)
    I am pretty sure Republicans have been saying regulations are bad and we have way too many of them for my entire life. How did that become âoeflood the zoneâ with regulations? I do not understand this timeline.
    • I am pretty sure Republicans have been saying regulations are bad and we have way too many of them for my entire life.

      They lie.

      How did that become Ãoeflood the zoneà with regulations?

      Anything is permissible in pursuing what they want.

      I do not understand this timeline.

      That is because you expect humans to act coherently. They don't. All humans are blind regarding Reality. As babies, we bump against Reality and realize that it can hurt. We then spend the rest of our lives trying to find the comfort and security of the womb, ignoring any detail that removes us from that spot. We are blinded by our own desires.

      And then, despite it all, we die. What fun life is.

  • The AI can probably get the regs 95% there and help weed out loopholes and ways to game the system.

  • First, they train the LLM with the many thousands of pages of federal regulations. Then, they use the LLM to 10x the number of pages.
    Shazam! They have more input to continue training, and the cycle repeats ad nauseam. Even the lawyer-bots won't be able to keep up.

  • Maybe this will mean somebody, anybody, will read a bill before voting on it. If so, the next-level idiocy of tasking Gemini to hallucinate regulations could literally prevent the next PATRIOT Act, the next DMCA, the next GENIUS Act, Inflation Reduction Act, etc.
  • Please oh please let this happen. As a veteran train wreck enthusiast and a top tier enjoyer of schadenfreud, I wish for this to happen. It can only make my doomscrolling better.

  • Presumably there are hardly any human staff left working at the DoT after Musk's DOGE cuts to staff numbers ? So amending a large number of regulations would be a really onerous task for the humans ...
  • and mandate Batmobiles with twin gas turbines for all!

  • It will include nonsensical hallucinations and random Nazi quotes.
  • Had the "Liberation Day" tariffs been drafted with a human, the penguins on South Georgia Island near Antarctica would have been left out. I look forward to other formerly overlooked places being included with this next round. How about rail service guidelines in Pitcairn Island or road construction regulations for Little Diomede Island? With AI, anything is possible.

The life of a repo man is always intense.

Working...