The Office of Legislative Counsel (OLC) of the U.S. House of Representatives has encountered an unexpected side effect of implementing generative AI: the flow of draft bills created by neural networks has surged sharply, adding extra strain on staff. Based on my information from current and former employees of the agency, the problem is systemic and requires an immediate response.
The key difficulty arises at the review stage. AI-generated documents are rife with references to nonexistent or outdated laws, and contain incorrect definitions and wording that distort the author's original intent. As a result, lawyers have to spend more time fixing such materials than preparing a document from scratch. Every inaccuracy is a potential delay in passing an initiative or, worse, a cause for future litigation.
Congresswoman Norma Torres has expressed concern about this, warning that over-reliance on AI increases the risk of errors and the appearance of wording that does not reflect the legislator's intentions. She emphasized: "This responsibility cannot be shifted onto a chatbot." One cannot disagree with that—delegating legal responsibility to an algorithm that does not understand the nuances of law is a dangerous precedent.
Official statistics do not yet allow an assessment of the scale of AI's impact on the OLC's workload. The head of the office, Warren Burke, in April 2025 linked the rise in requests to overall legislative activity and an increase in the number of amendments. However, as early as 2024, POPVOX experts noted that large language models had simplified the creation of pseudo-bills, and their flow to lawyers for review had noticeably increased.
Congress continues to adopt AI despite the risks
It is telling that the House of Representatives is not abandoning the technology. As stated by Aubrey Wilson, Director of Government Innovation at the POPVOX Foundation, work with Microsoft Copilot in the lower chamber continued in 2026, and starting in February, staff began training on this tool. She rightly emphasizes that AI should be implemented as an auxiliary tool that enhances human labor, not as a replacement for it.
The OLC itself is also actively exploring the technology. In April, Burke reported that the office, together with other legislative branch bodies, is considering possible scenarios for using neural networks, analyzing the associated benefits and risks.
It is worth noting that such problems are not limited to the United States. Recently, Guardian journalists found traces of AI hallucinations in a report for the Australian government, where the document contained references to nonexistent scientific publications. This confirms that the use of AI in the legal and governmental spheres requires an extremely cautious approach.
My analysis: The situation in the U.S. Congress is just the tip of the iceberg. We are witnessing a classic conflict between the speed of content generation and the quality of its verification. Until AI learns to check its own results for legal validity, its use in lawmaking will remain a risky experiment that could cost both legislators and citizens dearly.