On February 5, 2026, South Dakota state Rep. Al Novstrup stood on the House floor to argue for a bill expanding access to ivermectin and hydroxychloroquine. He is not a pharmacologist. He runs amusement parks. His research method? He asked Google Gemini to summarize the topic, then read the AI-generated summary aloud to his colleagues as supporting evidence.

"I am not qualified to know the answer," Novstrup told the chamber, "but here is some of the things I have learned by just doing some quick artificial intelligence." He then read Gemini output into the official legislative record of a state government.

South Dakota Searchlight investigated and found that AI has quietly crept into the entire 2026 legislative session. Lawmakers are using ChatGPT, Grok, Claude, and Gemini to draft bills, build arguments for floor speeches, refine talking points, and summarize complex policy issues. The 24-year veteran legislator compared it to using the Dewey Decimal System. That comparison tells you everything about the level of scrutiny being applied.

No Disclosure. No Rules. No Guardrails.

Here is the part that should concern everyone: there are zero requirements for lawmakers to disclose when they used AI to draft legislation or build arguments. No transparency rules. No auditing process. No quality checks on AI-generated content that ends up in the legislative record. A state representative can ask ChatGPT to write a bill, introduce it, and vote on it without ever telling voters that a chatbot helped write the law that governs their lives.

This is not just a South Dakota problem. It is a preview of every statehouse in America. If lawmakers in one state are doing this openly and casually, the safe assumption is that it is happening everywhere. Generative AI tools are free, fast, and persuasive. They are also confidently wrong on a regular basis. The news industry is littered with cautionary tales of factual errors, hallucinated sources, and context failures.

The Irony Is Almost Too Perfect

Just this week, a San Diego attorney was hit with $96,000 in sanctions for submitting AI-hallucinated court filings. The judge called it a "notorious outlier." Courts are actively punishing professionals who use AI carelessly. Legislatures? No consequences at all.

We have created a system where a lawyer can be fined six figures for letting ChatGPT write a brief, but a lawmaker can let ChatGPT write a bill that becomes binding law and nobody blinks. The same tool that hallucinates case citations is now hallucinating policy arguments that shape legislation. The difference is that bad legal filings get caught by judges. Bad legislation gets caught by the people living under it.

Novstrup says you always have to "be skeptical" of AI output. Sure. But skepticism requires expertise to evaluate what the AI got wrong. A lawmaker who admits he is not qualified to know the answer is using a tool famous for inventing answers to fill his knowledge gap. That is not research. That is outsourcing democratic governance to a language model that cannot tell the difference between a peer-reviewed study and a Reddit thread.

First reported by South Dakota Searchlight. Additional coverage from KOTA-TV and Dakota Free Press.