Congressional staffers and outside groups are increasingly turning to artificial intelligence tools such as Claude and ChatGPT to draft legislation, but the lawyers tasked with vetting that work say the technology is creating more problems than it solves. The House Office of Legislative Counsel, the nonpartisan office that turns policy ideas into precise statutory language, has been strained by a wave of machine-generated submissions. One person who has advised the office on the technology described the situation bluntly: the office spends “more time trying to fix AI-drafted legislation than it would take for them to draft it from scratch.”
The assessment was part of a detailed look at how AI is reshaping the legislative process in the House. It drew on interviews with eight current and former officials who work or have worked with the office. Many were granted anonymity because they lacked authorisation to speak on the record. Several congressional staffers involved in drafting said AI use is becoming routine in the House. “It’s absolutely terrifying,” one House staffer said, “and no one is thinking about it yet.”
Why the Office of Legislative Counsel matters
The Office of Legislative Counsel serves as the quality-control layer between a rough policy idea and an actual statute. When a member of Congress wants to introduce a bill, the office helps draft the precise legal language required for the measure to be considered by the House. “When we talk about Congress’s ability to legislate, we are talking about this office,” said Marci Harris of the POPVOX Foundation.
The volume of work is already enormous. In the 118th Congress, the office prepared 30,494 bills, which is 188 percent more than the 10,564 bills that were actually introduced. That statistic comes from testimony given in March by Warren Burke, the legislative counsel. The office also prepared 21,484 amendments, a 46 percent increase compared with the 114th Congress.
Burke’s testimony also revealed the resources available for this workload. In the first session of the 119th Congress, the office had 78 attorneys and 23 support staff. Attorney numbers are up 59 percent since the 116th Congress. Average experience has fallen 22 percent over that same period, and 58 percent of attorneys now have less than a decade of service. That evolving profile makes it harder for the office to absorb more complex or poorly drafted submissions without slowing down.
Where AI goes wrong in a bill
The errors described by those who work with the office are small but consequential. Wade Ballou, who led the office from 2016 to 2024, explained that AI struggles to distinguish between important legal mechanisms. It has difficulty telling whether a pot of money should be a “tax credit, tax deduction, tax exclusion or a grant.” Those distinctions affect how a law interacts with the rest of the tax code, and missing the nuance can create unintended consequences.
Definitions cause similar trouble. A bill that defines a state as only the 50 states can inadvertently exclude Washington DC and tribal nations from a federal programme. A person who works with the office offered that example, noting that such a small drafting choice can have a large impact on who receives benefits or protections.
Citations are the third major failure mode. Ari Hershowitz, a lawyer and technologist who has worked with the office, called incorrect statutory references “a guaranteed way of introducing thousands of bugs into our legal system.” When a bill cites the wrong section of the US Code, courts and agencies may interpret the law in ways that Congress did not intend.
Daniel Schuman of the American Governance Institute offered a broader warning. “AI is good for a lot of stuff,” he said, “but it’s not capable of drafting legislation that you would want to enact into law.”
This is not the first time that official pipelines have buckled under machine-generated submissions. In August, a security researcher filed 55 software vulnerabilities in a short period, and 54 of them turned out to be invented by AI. The incident showed how quickly automated content can overwhelm systems designed for human work.
Two different growth figures
Burke’s testimony from last year cited a 72 percent rise in requests to the Office of Legislative Counsel. That figure covered the first 60 days of the 119th Congress, comparing 5,623 requests against 3,271 in the same window of the previous Congress. It suggested that the AI effect was immediate and steep.
Burke’s most recent testimony reports a different rate over a longer window. Across the first 13 months of the 119th Congress, the office received 23,239 requests. The comparable figure for the 118th Congress was 21,979, a rise of 5.7 percent. The difference between the two numbers matters: a short-term spike can be driven by particular legislative priorities, while a longer trend can show a steadier increase in demand.
On AI specifically, Burke was direct in March. “In the near term, we are bracing for AI to expand our workload,” he wrote. He added that text produced by outside entities, whether human or computer, often takes significantly longer to review and revise. That suggests the office is not simply handling more requests; it is handling requests that require more effort per item.
The office is testing its own AI
While outside AI tools are creating headaches, the Office of Legislative Counsel is also exploring how AI might help its attorneys. The office has organised a group of staff to evaluate Microsoft Copilot, as authorised by the House. The stated aim is improving efficiencies such as legal research, according to Burke’s testimony.
The office already runs one tool of its own. The Comparative Print Suite uses natural language processing to show exactly how a bill would change the US Code. Hershowitz described it as the first AI tool in the House. It returns an error when it cannot place a change with confidence. “There’s no hallucinations,” he said. That reliability is important in an environment where a hallucinated citation could derail a law.
Charlotte Lee of Granicus helped design that tool, and she described a striking asymmetry in how AI is used. “Members and interest groups get to write things with AI, but then leg counsel themselves can’t use [commercially available] AI,” she said. “There’s a complete incongruency.”
What has not been established
Despite the rise in AI-generated drafts, no bill voted on by either chamber has turned out to be entirely AI-crafted. The evidence so far concerns AI-drafted bills arriving at the office, not statutes leaving Congress. That distinction is important because the legislative process includes human review at multiple stages, even if that review is strained.
Disclosure is patchy by nature. One staffer noted that people who use AI tend not to talk about it, and that the topic still feels taboo. Labelling rules elsewhere have landed unevenly. Under the EU AI Act, an AI-written article needs no label while a proofread email does. That inconsistency reflects how rapidly and unevenly AI norms are developing.
One incident involving AI in the House has become public. Representative Anna Paulina Luna drew criticism in June over an amendment summary generated with Claude. She said she had not used AI to write the amendment itself. “Not a shocker. Most staff use it,” she said, adding later that she had told her staff to double check. She later deleted one post on the episode and replaced it.
Two members of the Committee on House Administration gave statements. “AI cannot replace the skilled and dedicated professionals in the Office of Legislative Counsel,” said Representative Joe Morelle. Representative Norma Torres said responsibility for introduced legislation rests with members and their staff, and “cannot be outsourced to a chatbot.”
The wider AI push on Capitol Hill
Congress has been circling AI from several directions this year. It spends nine of ten AI dollars with OpenAI. House Democrats want answers from OpenAI and Anthropic on rogue agents, after incidents in which autonomous AI systems took unexpected actions online. In July, a rogue agent breached Hugging Face, a popular platform for machine-learning models.
Legislation is moving as well. A bill introduced by Ted Lieu and Jerry Moran would create an emergency kill switch for AI systems. The proposal is one of several efforts to impose rules on advanced AI as it becomes more capable.
Ballou sees a narrower role for the technology. AI can help lawmakers process information, he said, but it cannot replace the art of drafting. Hershowitz expects the volume problem to worsen either way. “You had a problem that was already at 10,” he said. “Now you multiply it by 100.”
Source: TNW | Government-policy News