I opened the staffer’s message at 2 a.m.; the subject line read, “It’s absolutely terrifying.”
It was a short chain: a deleted tweet, an edited bill, and a chatbot called Claude sitting in the middle.
You get a chill when the words you scroll past tonight can become the rules people live under tomorrow.
I’ve tracked messy policy fights and sloppy memos. I’m telling you this because the signals are consistent and getting louder: something that started as convenience is seeping into the spine of legislation, and you should care.
A staffer posted a deleted tweet linking Claude to a bill
Representative Anna Paulina Luna publicly flagged a staffer’s use of Anthropic’s Claude in June, then removed the clarity from the record.
The timeline is odd: Luna posted apparent evidence that Claude had helped draft language, later replaced that claim with a softer note that the chatbot produced only a summary. The original tweet was deleted. That ambiguity is exactly the problem—when public threads and private drafting intersect, no one gets a clean audit trail.
Is AI being used to write bills in Congress?
Yes—at least according to anonymous congressional staffers who spoke to Politico and the mix of signals from lawmakers. They say outside groups and some Hill staff are turning to tools such as Anthropic’s Claude and OpenAI’s ChatGPT to generate legislative language directly.
Requests for legislation jumped 72% in 2025
The Office of Legislative Counsel, led by Warren Burke, told the House that requests for drafting were up 72% in 2025 compared with the same stretch in 2023.
That stat matters because the OLC is the gatekeeper for precise, technical language. Staffers told Politico the office now spends more time untangling AI-drafted text than it would take to write clear language from scratch. I’ve seen that torque before: what begins as a time-saver ends up costing time—and trust.
How does AI affect legislative drafting?
AI gives more people the confidence to produce jargon-heavy drafts, but confidence is not the same as competence. When chatbots generate dense legal verbiage, mistakes hide easily; a misplaced clause can change civil rights, budgets, or regulatory scope overnight. In practice, AI-drafted bills can be like an unvetted recipe—one missing ingredient spoils the outcome.
Outside groups are sending whole drafts from chatbots
Staffers report that advocacy groups, contractors, and lobbyists are increasingly pasting chatbot output straight into bill text.
That means language originating from Claude or ChatGPT can enter the legislative pipeline without the traditional legal review. The Office of Legislative Counsel then has to rewrite or repair it, which wastes scarce institutional time and increases the chance that something slips through. These drafts can function like a Trojan horse: a tiny phrasing change can carry large legal consequences embedded at first glance.
Can AI-drafted laws be trusted?
Trust is a legal question, not a marketing one. Machines do patterns, not policy judgment. If a court later interprets a statute that was originally assembled by an AI, the stakes become constitutional and practical—ambiguity begets litigation, and litigation creates precedent.
Here’s what matters next: Congress still relies on human lawyers in the OLC to stamp and clean text, but volume and outside pressure are changing that workflow. Regulators, tech companies, and legal drafters will argue about responsibility, but the immediate friction is operational: more fixes, longer waits, and more opportunities for errors that matter to millions.
I can show you where to watch the threads: follow Politico’s reporting, read Warren Burke’s House statement, and track the public posts from lawmakers who raise the alarm. Watch Anthropic and OpenAI, whose tools are named in these reports, and expect more hearings and memos.
Who will answer for a comma drafted by an algorithm when it rewrites rights or budgets—Congress, the company, or the staffer who hit send?