Earlier this week, Owen Dahlkamp at Politico published a report about how a flood of legislative proposals, likely driven by AI usage, has begun to overwhelm the House Office of Legislative Counsel (HOLC) of the U.S. House of Representatives, forcing the organization to spend increasing amounts of time fixing errors and rewriting proposed legislation.
According to the article, the allegations come from four anonymous staffers with the organization. They claim that AI proposals have greatly increased the number of proposals submitted to the HOLC and also increased the number of errors and issues that need to be fixed.
The HOLC provides legislative drafting services to various committees and members of the House of Representatives. The service is non-partisan and confidential, and aims to help draft legislation that is “clear, concise and legally effective.”
As of April 2025, the office employs some 73 attorneys and 23 support staff. In the first sixty days of the session, the office saw a 72% increase in the number of requests received during the first sixty days of the session versus two years ago.
According to the anonymous HOLC staffers, much of that increase is being driven by AI. Both staffers serving members and outside agencies have made increasing use of AI to draft legislation, leading to a significant rise in both the number of bills being proposed and the number of corrections those bills need.
This risks overtaxing a little-known but extremely-important office in the nation’s legislation. However, it’s far from the only piece of human infrastructure cracking under the weight of AI. Those institutions are fracturing all around us.
When AI Meets Human Institutions
One of the most obvious examples of these issues is how AI has impacted the legal system. AI has spurred a massive spike in the number of unrepresented (pro se) filers. According to one paper, pro se filings in FY 2025 is almost double the pre-AI average.
Not only is there more pro se cases being filed, there is more activity within those cases. This is leading to serious issues as judges and lawyers alike are trying to address the spike in activity. These filings often have serious errors and other issues that need to be addressed, delaying cases and slowing down resolution times.
Of course, it isn’t just pro se litigants making dubious use of AI. Lawyers and expert witnesses have as well. Still, it’s the spike in pro se litigants, powered by AI, that has caused the most alarm.
But it’s not just the courts facing these issue. On the technical side of things, bug bounty programs, which are essential for detecting and eliminating dangerous security flaws, have also become similarly overrun.
Apple has faced struggles weeding out AI-generated bug reports, GitHub has had to implement a teired bug bounty system and the cURL project had to drop theirs altogether. Even Linux itself has seen reports rise 76% year-over-year.
All of these cases have one thing in common. They are all situations where AI has made it easy to generate something that a human then has the responsibility of validating and/or fixing. AI greatly lowered the barrier of entry, but not the time required to process the submission.
Many of these institutions were already fragile. Now, in many cases, they are simply breaking.
The Challenge with Legislation
The challenge with legislation is that, even before AI, it was often unclear who drafted a particular piece of legislation and how they did it.
Though members of the House and Senate are listed as the bill’s authors, they rarely are in any meaningful sense. At the bare minimum, it’s usually their staff and the OLC that does the majority of the drafting.
However, even that may not be the actual origin. Lobbyists and other outside entities often provide draft legislation that makes it into the final bill. With a lot of legislation, there is little to no indication of who wrote, how it was written and how it evolved as it made its way through the system.
Such a system is especially ripe for AI misuse. That’s because no one really has to take responsibility for the AI text and any problems with it.
With a pro se litigant using AI, at least the judges and attorneys can hold them to some account. But with legislation, there could be multiple lobbyists, staffers across multiple offices, committees and more. The only person we can say definitively didn’t use AI is probably the person (or people) listed as the bill’s author.
This creates a real problem when trying to create or implement any kind of AI policy. There’s simply no way to enforce it.
Though legislation is definitely a space that is far too important to leave vulnerable to AI errors and issues, there’s almost no way to keep it out. The only solution we have right now is for humans at the OLC to do the thankless work of cleaning up the mistakes and errors.
But that is likely an untenable solution over the long haul.
Bottom Line
When it comes to authorship, the legislative process in the United States is easily one of the most bizarre and unusual. It’s a situation where the named author had very little to do with the drafting and the actual work is done through a mishmash of staffers, lobbyists and lawyers, who all work in a completely opaque environment.
This system has long been ripe for abuse. AI just adds another potential means for that abuse to take place.
But AI is a particularly dangerous form of abuse. It stresses the already fragile system that turns ideas into law. It slows down the already sluggish legislative process. That’s because AI is great at producing a large quantity of material, but not as good at producing high quality material.
With legislation, quality has to be the focus. The humans who are tasked with ensuring those quality guidelines are met are a fragile and underappreciated part of the process that are now being drowned in AI-generated legislation.
There is no easy solution here. Without a fundamental shift in the way legislation is drafted, this problem will only get worse.
However, this is one institution that we can not afford to lose. For better or worse, it is at the very heart of our government. I just don’t know if there’s enough will to ensure that it doesn’t buckle under the weight of AI.
Want to Reuse or Republish this Content?
If you want to feature this article in your site, classroom or elsewhere, just let us know! We usually grant permission within 24 hours.