AI slop is flooding the House bill-drafting office
The House's statute-writers now spend more time unpicking chatbot bills than they would writing the laws themselves.
The House Office of Legislative Counsel is being buried under AI-generated bill drafts — many of them wrong — and the lawyers who turn policy ideas into statutes say the cleanup is slower than starting from a blank page. Politico, citing interviews with eight current and former officials, reports that congressional offices and outside groups are feeding Claude and ChatGPT the job of writing legislative language. One person who has advised staff on the tools said the office now spends "more time trying to fix AI-drafted legislation than it would take for them to draft it from scratch." Townhall's write-up of the same reporting lands in the same place: the bottleneck is not imagination, it is quality control.
This is a different problem from aides using chatbots to sort mail or draft a floor speech, which we covered earlier this month — Chatbots are doing Congress's work with little oversight. Speeches can be embarrassing. Statutes can be sued. Wade Ballou, who ran the office from 2016 to 2024, told Politico that models miss the distinctions that actually decide who gets paid — whether a pot of money is a tax credit, a deduction, an exclusion, or a grant. People who work with the office described drafts that define "state" as the 50 states and quietly drop D.C. and tribal nations from federal programs. Ari Hershowitz, a lawyer who has helped the office build drafting tools, called bad citations "a guaranteed way of introducing thousands of bugs into our legal system."
The volume was already ugly before the chatbots showed up. In the last Congress, Legislative Counsel's 61 attorneys and 19 support staff prepared more than 30,000 bills — nearly three times as many as members actually introduced. In the first 60 days of the current Congress, the office took 5,623 requests, up 72 percent from two years earlier, according to office head Warren Burke's testimony to House appropriators. Burke, who declined to comment for the Politico story, told a March hearing that text from outside entities, "whether a human or a computer, often takes significantly longer to review." No bill that has passed either chamber has been shown to be entirely written by AI. The risk is the ones that skip the review because the review now takes too long.
The take writes itself, and it is not "Congress should ban ChatGPT." Members and lobbyists get to generate text at machine speed. The civil servants who have to make that text mean something in the U.S. Code do not get matching tools, a gap Charlotte Lee of Granicus called "a complete incongruency." Daniel Schuman of the American Governance Institute put the capability question more bluntly: the models are not capable of drafting legislation you would want to enact. Hershowitz's version of the same math: the office already had a problem at 10, and AI multiplies it by 100. If the only office that can still catch a hallucinated statute citation is slower than the people generating the errors, the next failure will not look like a bad press release. It will look like a law.
What to watch: whether members start introducing unreviewed AI drafts rather than wait in the Legislative Counsel queue.
If a chatbot can write a bill faster than a lawyer can fix it, should that bill be allowed on the floor? Tell us in the comments.
Sources: Politico · Townhall · Techmeme · Warren Burke March testimony
