Congress Members Raise Concerns Over AI Errors in New Legislation — Politico

Members of the OLC in the U.S. have stated that they are receiving an increasing number of new bills showing signs of AI usage.
Such documents contain errors, incorrect formulations, and significantly complicate the work.
These instances are becoming more frequent and may pose risks in the future.

Legislators in the U.S. are using AI to draft texts for new bills, which poses long-term risks and creates additional burdens. This was reported by Politico, citing its sources.

This viewpoint was expressed by several current and former staff members of the Office of Legislative Counsel of the U.S. House of Representatives (OLC). According to them, it takes more time to review and correct one bill created with AI than to write a document from scratch.

Moreover, this approach threatens legal challenges and other consequences in the future due to incorrect, and sometimes erroneous, formulations.

Artificial intelligence cannot replace qualified and dedicated professionals in the Office, who help the House of Representatives perform its most important work, — stated Congressman Joe Morelle.

The reason AI struggles with such tasks lies in the multitude of nuances that must be considered when drafting a bill. For example, the difference in defining a tax credit, tax deduction, tax exemption, and grant.

Additionally, AI incorrectly cites previously passed laws, sometimes invents them, and confuses definitions and government structures.

At the same time, the number of new bills is constantly increasing. It is expected that AI will continue to drive this trend, which will likely complicate the work of the OLC and lead to delays in the amendment process and bill preparation, Politico concluded.

The situation is exacerbated by the fact that employees who use AI to write documents do not disclose this, the publication noted.

One sign that such tools are being increasingly relied upon is the weight of the documents. The amount of text, often excessively inflated with unnecessary structures, in new bills is only growing, the OLC noted.

At the same time, despite ongoing efforts to improve workflows, the Office cannot use commercial AI to check and edit bills.

AI in Politics: The Problem is Becoming More Relevant

In July 2026, New Brunswick Legislative Assembly member Bill Oliver stated during a speech:

Here is a more natural, smoother version of this section that sounds like legislative speech rather than a set of bullet points.

Moreover, similar insertions occurred at least twice more. Later, he admitted that he used AI while preparing the speech and simply did not proofread the final version before printing.

And similar cases are happening everywhere. For instance, the European Parliament is developing an internal AI system, EPGenAI Hub, which will serve as a gateway to commercial models. The project's goal is to minimize the use of third-party solutions and poor processing of bills.

The Parliament has not avoided mistakes related to AI and misleading content generated by AI. Members of the European Parliament must set an example, as the transparency rules outlined in the AI Act become applicable, — stated MEP Brando Benifei.

Among other cases directly in the White House, one can highlight the office of Anna Paulina Luna, a member of the House of Representatives. According to WSJ, before the amendment to the NDAA defense bill, her team added a note, the beginning of which contained an unedited fragment: Claude responded.

As it turned out, an employee copied Claude's response along with the timestamp and service line directly into the public record of Congress. Luna confirmed the use of Claude, although she emphasized that AI only made the annotation and technical editing, not the text of the amendment itself.

Earlier, we also reported that in Utah, a lawyer lost his job for using AI to draft complaint texts. Meanwhile, the chatbot referenced non-existent precedents to support the claims.