Congress Is Using AI to Write Bills — and the Results Are a Mess

The office that writes laws for Congress is being flooded with AI-generated drafts full of mistakes, according to a report published by Politico on August 17, 2026. The lawyers who fix these drafts say they now spend more time correcting them than they would spend writing the bills from scratch, sources told reporter Owen Dahlkamp. (Politico)

More congressional offices are turning to publicly available AI tools, specifically ChatGPT and Claude, to produce legislative proposals, Politico reported. The result is a growing pile of defective draft bills arriving at the Office of Legislative Counsel, or OLC. This is the nonpartisan office responsible for turning policy ideas into legally sound language that can actually become law. (Engadget)

The errors are not small. Wade Ballou, who led the OLC for nearly a decade until 2024, told Politico that AI cannot tell the difference between a tax credit, a tax deduction, a tax exclusion, and a grant. These are not just different names for the same thing. Each one has different legal and budgetary consequences under federal law. A model that picks the wrong one has not drafted a bill with a minor flaw. It has drafted a different bill.

One recurring error reported by Politico is that AI drafting tools sometimes define "state" to mean only the 50 US states. That would leave out the District of Columbia and tribal nations, meaning they could be excluded from federal programs without anyone intending it. A lawyer who worked with the OLC told Politico that AI also cites previous laws in draft bills that do not exist, or that do not say what the draft claims they say.

Beyond the technical mistakes, Politico reported a subtler concern. Congressional staffers who use AI to draft their bills are not as deeply familiar with what their bills are actually about. The OLC's work has always depended on a back-and-forth process in which staffers explain their policy goals and the legislative counsel translates those goals into precise legal text. When the staffer's starting point is a machine-generated draft, the gap between the proposal and its author widens. The OLC's job shifts from collaboration to detective work.

The OLC is not rejecting AI entirely. A team within the office built an internal tool called the Comparative Print Suite, which helps staffers see how a proposed change would alter current law. The tool has a built-in safety measure: when it cannot figure out where a change should go, it returns an error instead of guessing. That choice keeps made-up text from sneaking into draft bills, a problem the office has apparently seen plenty of from consumer AI tools.

The contrast between the two approaches is striking. General-purpose AI tools, trained on text from across the internet, are designed to produce smooth, confident writing regardless of whether that writing is legally correct. The Comparative Print Suite, built specifically for the legislative environment it serves, limits its output to what it can verify and simply stops when it cannot. One approach prioritizes producing something. The other prioritizes producing something right.

Anyone who works with technology will recognize this pattern from other areas. The same qualities that make AI useful for writing marketing copy or summarizing documents — its fluency and speed — become a problem when the output must follow precise rules with no room for confident-sounding mistakes. Legislative drafting is more like writing computer code than writing an article. A citation that sounds right but is wrong is not a style problem. It is a functional defect. The difference is that bad computer code usually fails quickly, where it is easy to spot. A bad legal citation can travel through the legislative process until someone catches it, or does not.

In my view, the lesson here is not that AI is unsuited to legislative work. The Comparative Print Suite shows that a tool built carefully, within the constraints of its specific field, can genuinely help. The problem is the gap between what general-purpose chatbots can produce and what writing laws actually requires, and the assumption by some congressional offices that the first is good enough to replace the second. The OLC's lawyers are now paying the price for that gap, one fabricated citation at a time.