The House Office of Legislative Counsel, the 61-attorney, 19-support-staff shop that turns member ideas into statutory language, is now spending “more time trying to fix AI-drafted legislation than it would take for them to draft it from scratch,” according to an adviser to office staff quoted in Politico’s August 18 report.

Eight current and former officials told Politico that congressional staffers and outside advocacy groups are feeding prompts into ChatGPT and Claude and dropping the output directly into legislative requests. Volume tells the rest. OLC logged 5,623 requests in the first 60 days of this Congress, up 72% from two years earlier.

The failure modes are specific, not theoretical. Wade Ballou, the office’s former head, told Politico the models can’t reliably distinguish a “tax credit, tax deduction, tax exclusion,” and have defined “state” in ways that exclude DC and tribal nations. A lawyer who worked with the office flagged AI drafts miscitng prior statutes, the kind of error that surfaces years later in litigation.

The tell sometimes surfaces earlier. In June, Rep. Anna Paulina Luna’s office posted an amendment summary containing the raw string “Claude responded:”. Luna said the tool was used on the summary, not the amendment, and offered the broader defense: “Not a shocker. Most staff use it.”

OLC disclosed in March that it had formed an AI working group and built an internal tool called Comparative Print Suite to visualize how proposals would change current law. It’s a defensive posture, not an answer.

One House staffer’s assessment, in Politico: “It’s absolutely terrifying, and no one is thinking about it yet.” The institution debating whether to regulate frontier AI is quietly running a live experiment on itself, and the quality-control layer is 61 lawyers deep.

Sources

Sources