Artificial intelligence delusions are seeping into one of the most sensitive decision-making arenas in the United States, as generative AI tools begin to be used to prepare legislative texts that reach Congress with legal and drafting errors.

According to a report published by Politico, the Office of the Legislative Counsel for the U.S. House of Representatives is facing an increasing flow of bills generated by artificial intelligence, which requires additional time from its lawyers to review and rewrite them.

Many drafts and frequent errors

The issue was based on interviews with 8 current and former officials who worked in or dealt with the Office of Legislative Counsel, who reported that the use of artificial intelligence to prepare bills has become more widespread within the House, as some offices and external groups rely on tools such as ChatGPT and Claude to produce legislative language itself.

The problem is not just linguistic errors; rather, some drafts arrive loaded with errors that can be more complex, including incorrect references to laws or legal articles, flawed legal definitions, and poor drafting.

Artificial intelligence’s main advantage is its ability to produce large amounts of text quickly. But this same advantage becomes a problem when the texts require accurate legal review.

According to sources who spoke with Politico, the lawyers in the Office of Legislative Counsel are forced to spend more time reviewing and rewriting some AI-generated proposals. In certain cases, correcting the text can take longer than drafting a new version from scratch.

The danger of these errors does not stop at the review office. If an incorrect legal reference or inaccurate definition is entered into a legislative draft, it could delay the approval process or create legal problems later on.

The report indicated that the mistakes can have repercussions beyond Washington, including the potential for lawsuits due to reliance on incorrect laws or use of incorrect legal definitions.

This does not mean that AI is prohibited from being used in legislative work, or that every bill prepared with its help is flawed. The problem lies in relying on the model to generate legal language without specialized human review.

Reports quoted Daniel Schuman, executive director of the American Governance Institute, as saying that AI is useful in many areas, but it is not ready for drafting legislative language that users want to pass into law.

Even understanding the idea can become a problem

The risks don’t just relate to the drafting of the text alone, according to Politico. Some people who submit AI-generated drafts may struggle to explain the precise political or legal outcome that they want the text to achieve.

This adds another problem: when a legislative idea is turned into a full text with the click of a button, it can look more mature than it actually is. The underlying idea still needs to be specified and reviewed by its authors.

The issue underscores that using AI in government institutions doesn’t eliminate human responsibility for outcomes.

The US Department of Justice stressed that the primary responsibility for bills submitted lies with members of Congress and their staff, not a chatbot.

Despite the growing prevalence of generative tools, there are currently no confirmed cases where the House or Senate has voted on legislation drafted entirely by AI.