The legal service of the U.S. House of Representatives (OLC) has faced a new reality: the flow of draft bills created with generative AI has surged sharply, placing serious strain on the system. Based on my data, drawn from confidential discussions with current and former agency employees, the problem lies in the quality of such documents.
AI generation, for all its apparent efficiency, produces texts with critical flaws: incorrect references to current regulations, distorted legal definitions, and wording that often fails to reflect the legislator's true intent. As a result, lawyers have to spend more time fixing such "semi-finished products" than writing a document from scratch. Each such inaccuracy is a potential delay in passing an initiative or, worse, a trigger for future litigation.
Congresswoman Norma Torres expressed concern that over-reliance on AI increases the risk of legally untenable texts. "This responsibility cannot be shifted onto a chatbot," she emphasized, and I fully agree with her.
Official statistics do not reflect the scale
Official data does not allow for an accurate assessment of AI's impact on lawyers' workloads. In April 2025, OLC head Warren Burke attributed the rise in requests to overall legislative activity and an abundance of amendments. However, as I noted in my analytical reviews, as early as 2024, POPVOX experts warned that large language models had significantly simplified the creation of pseudo-bills, inevitably increasing the flow of junk documents for review.
Congress is not abandoning AI
Despite the chaos with drafts, the House of Representatives is not scaling back AI adoption. In 2026, work with Microsoft Copilot in the lower chamber continues, and since February, employees have been actively trained to use this tool. Aubrey Wilson, director of government innovation at POPVOX Foundation, rightly emphasizes: AI should be an auxiliary tool that enhances human labor, not a replacement for it.
The OLC itself is also exploring the capabilities of neural networks. Burke reported that the service, together with other legislative branch structures, is analyzing use cases, benefits, and risks. This is the right approach, but it takes time, which legislators, as always, lack.
My verdict: the situation in Congress is a vivid example of how a technology designed to speed up processes creates new bottlenecks. AI tools are not yet capable of replacing legal expertise and only add work to those responsible for the quality of laws. This is a lesson for all industries where accuracy is critical: before delegating tasks to neural networks, a verification system must be built, otherwise "time savings" will turn into multiplied losses.