Congress’ Office of Legislative Counsel fielded 72% more requests for legislation in 2025 than it did two years prior. The figure was disclosed by Warren Burke, the office’s head, during remarks to the House last year. Four congressional staffers, speaking anonymously, now believe they can explain part of that surge.
The staffers, who by their own account work on drafting bills, talked to a story that ran Monday. Their allegation: text produced by AI is finding its way into legislation at a worrying pace.
The worry has a concrete basis. Bills that make it through become federal statutes — part of the bedrock of American life.
“It’s absolutely terrifying,” one of them said.
The people writing the language often aren’t in the building
According to one of the anonymous staffers, much of the problem originates outside Congress entirely. Individuals and organizations are “going to Claude or ChatGPT to draft the legislative language itself,” the person said.
That detail deserves a moment. The picture isn’t a tired aide polishing a paragraph after midnight — it’s outside groups delivering machine-written statutory language and leaving the professionals to untangle it.
Untangling is where the cost lands. The Office of Legislative Counsel is staffed largely by lawyers who draft and format laws. That same staffer said the office burns “more time trying to fix AI-drafted legislation than it would take for them to draft it from scratch.”
Framed as a productivity story, the math runs backwards. A tool marketed on time savings is manufacturing extra work for the one office with no ability to turn it away.
We’ve seen this exact failure before, in court filings
Assuming the account holds up, AI’s role in lawmaking looks a lot like its role in litigation. It emboldens more people to attempt documents written in dense, technical language that leaves no room for error. And it piles work onto whoever has to read those documents afterward.
Statutory language is a poor match for a system whose default setting is fluency. Precision is what keeps a statute standing, and prose that reads like law is a very different thing from prose that operates as law.
The Luna incident, and the tweet that disappeared
Back in June, in an episode that was both odd and hard to pin down, Florida Republican Representative Anna Paulina Luna shared what looked like proof online that a staffer had used Anthropic's Claude chatbot to draft legislation.
She subsequently appeared to acknowledge that the chatbot had at minimum been used to edit a bill.
Then the tweet vanished. In its place came another one stating the chatbot had done nothing more than summarize the bill.
Drafted, edited, summarized — three separate accounts of one bill, each narrower than the last, with the paper trail thinning at every step. It’s the entire problem compressed into a single episode: no one outside the office can determine which version is accurate, and nothing obliges anyone to say.
Anonymity on the staffers’ part caps how hard any of this can be pressed. Still, the 72% number is on the record, delivered by the person running the office that absorbs the workload, and it leans the same way the staffers do.