Congressional staffers and outside groups are dumping AI-generated bill text into the House Office of Legislative Counsel faster than its lawyers can clean it up, Politico reported this week after interviewing eight current and former officials tied to the office. OLC attorneys now spend more time debugging chatbot drafts than they would spend writing the legislation themselves.

The scale is unforgiving. In the 118th Congress, OLC’s 78 attorneys and 23 support staff prepared 30,494 bills and 21,484 amendments. Legislative counsel Warren Burke has testified that requests are up 72% year-over-year. One adviser described colleagues “going to Claude or ChatGPT to draft the legislative language itself,” then routing the output to OLC for cleanup.

The errors are the kind that matter: wrong tax terminology, flawed definitions of “state,” miscitations of existing statute. Ari Hershowitz, a lawyer familiar with the office, called it “a guaranteed way of introducing thousands of bugs into our legal system.” Wade Ballou, who ran OLC from 2016 to 2024, put it more quietly: “AI is going to miss those nuances.”

Formal guardrails exist and aren’t enforced. Both chambers have cleared ChatGPT, Copilot, and Gemini; the House also cleared Claude and bought 6,000 Copilot licenses last year. House guidance bars using AI to finalize legislation, generate deepfakes, or ingest constituent data. The Washington Post couldn’t find a single disciplinary case brought under it.

In June, Rep. Anna Paulina Luna (R-Fla.) filed an amendment summary that opened with pasted Claude output. Asked about it, she shrugged: “Not a shocker. Most staff use it.”

OLC’s answer so far is an internal AI working group and a homegrown “Comparative Print Suite.” Neither slows the inbound volume. The bottleneck between an idea and a US statute is being asked to absorb chatbot throughput at chatbot speed, on 101 people’s time.

Sources

Sources