The U.S. Congress has a growing problem that isn't just about partisan gridlock. It’s about a flood of legislative proposals written by machines that don't understand the law.
Lawyers at the House Office of Legislative Counsel (OLC) are currently buried under a mountain of AI-generated drafts. These aren't just minor typos or formatting issues. They are documents "rife with errors" that force legal experts to waste more time correcting them than it would have taken to write the bills from scratch.
The Myth of the Automated Lawmaker
Lawmakers and their aides are turning to tools like ChatGPT and Claude to draft statutory language. It’s tempting. It’s fast. It’s free. But in the world of federal law, speed is often the enemy of precision.
Drafting a bill isn't just about stringing sentences together. It requires deep, granular knowledge of the U.S. Code and the specific, often boring, nuances that determine how a law functions. If you're a congressional staffer tasked with writing a bill, you aren't just drafting a proposal; you're building a foundation for potential litigation, regulatory oversight, and massive fiscal shifts.
The fundamental issue is that LLMs don't "know" the law. They predict the next likely word in a sequence based on training data. They don't grasp the legal distinction between a "tax credit," a "tax deduction," and a "grant." If an AI classifies a funding stream incorrectly, the entire purpose of a bill can be undermined before it even reaches a committee floor.
Why Small Mistakes Cost Millions
Consider a specific technical detail: the definition of a "state." In federal statutes, that term often needs to account for territories, tribal nations, and the District of Columbia. A human drafter trained in the OLC’s rigorous standards knows to define these terms explicitly to avoid unintended exclusions. An AI might simply assume the 50 states, effectively nuking federal funding for millions of citizens overnight.
Then there’s the issue of citations. AI models are infamous for "hallucinations"—they’ll cite a U.S. Code section that sounds official but simply doesn't exist. When these errors land on the desks of the 61 lawyers and 19 support staff at the OLC, they have to treat the draft like a hazardous waste spill. They have to trace every claim, verify every citation, and fix every logic gap.
The Productivity Trap
The irony is thick. The goal of using AI was likely to increase legislative output, especially as the number of requests to the OLC has spiked—rising 72% in just two years. Lawmakers want to move fast, but by submitting AI-generated "slop," they are actually clogging the system.
It’s also eroding the legislative process itself. When a staffer uses a chatbot to write a bill, they aren't forcing their brain to engage with the policy. They aren't grappling with the competing interests or the practical outcomes of the law. They're just pushing a button. If you can’t explain the purpose and the mechanics of your bill without asking an AI to summarize it for you, you don't actually understand the legislation you're trying to pass.
What Happens Now
The OLC is trying to fight back with their own tools, such as the "Comparative Print Suite," designed to help staffers visualize how their proposals interact with existing laws. Unlike the general-purpose chatbots, these internal tools are designed to flag errors rather than just hallucinate more text.
But this isn't just a technology problem; it’s a culture problem. As long as congressional offices prioritize high-volume output over deep, substantive work, the flood of low-quality, AI-drafted proposals will continue.
If you’re watching the legislative process, don’t be surprised when you see more "clerical errors" in major bills. It’s the result of an office culture that’s trading craftsmanship for convenience. Legislation is the bedrock of our system. When you start building that bedrock with software that doesn't understand the law, you aren't just creating a mess for lawyers to clean up. You're risking the stability of the entire legal framework.
Stop looking for the easy way out. Real legislative drafting requires a human eye, a legal mind, and a deep, annoying attention to detail that no chatbot can replicate.
