(SeaPRwire) – By: Gavin Thorne
There is a quiet crisis brewing inside the House Office of Legislative Counsel. It is not about ideology. It is about who gets to write the law and what happens when nobody checks the work of a chatbot.
The numbers are stark. In the first 60 days of this Congress, the OLC received 5,623 legislative requests. That is a 72 percent increase compared to the same window two years ago. The office has 61 attorneys and 19 support staff. Nobody is surprised that the workload is crushing. But the new pressure does not come from harder bills. It comes from AI-generated drafts arriving at the OLC with incorrect statutory references and confused legal definitions.
Lawyers are now spending more time trying to fix AI-drafted legislation than it would take to draft it from scratch. One person who advises congressional staff on technology put it plainly. This is not a margin problem. It is a structural one. When a chatbot defines “state” in a way that could exclude Washington, DC or tribal nations from a federal program, you are not looking at a minor formatting error. You are looking at a bill that could silently carve out entire populations from benefits or protections without any legislator noticing.
What happened in June should have been the final warning. Republican Representative Anna Paulina Luna’s office published an amendment summary that read “Claude responded:” right in the text. She said AI had been used on the summary rather than the amendment itself. Most staff use it. That casual admission from a sitting member of Congress says more about the institutional culture than any policy memo ever could. The problem is not that AI is bad at law. The problem is that AI does not know the difference between a useful drafting aid and a source of legal authority.
Canada already went through this last year. New Brunswick lawmaker Bill Oliver read leftover chatbot instructions aloud in the legislature. The speech included the line: “Here’s a more natural, flowing version of that section that reads like a legislative speech rather than a series of short points.” The chamber heard it. The internet recorded it. No one resigned. Nobody was disciplined. The system absorbed it and moved on because the alternative is admitting that the entire process of how laws are written is about to be disrupted in ways that oversight bodies are not built to catch.
The OLC knows this. It formed a working group in March to examine AI uses for legal research. It has already introduced specialized technology to help its own lawyers draft legislation. That is the institutional reflex. Adapt or be bypassed. But adapting an office that is already drowning in requests is not the same as solving the deeper question of why AI-generated bills are reaching the drafting stage in the first place.
The answer is simple. Generative tools are fast. They are cheap. They produce text that looks like law even when it is not law. A staffer under pressure can run a policy idea through ChatGPT or Claude in five minutes and get back a plausible-looking draft. The draft will contain correct citations half the time. The other half of the time it will contain hallucinated ones. The OLC is the safety net. And the safety net is full.
This is not a technology story. It is a delegation story. Congress is handing its most consequential function to algorithms and calling it efficiency. The legal community inside the OLC is doing what it can to clean up the mess. That work will keep growing. The real question is whether anyone in leadership is willing to set a rule that says an AI draft cannot be introduced without human review before it ever reaches the floor.
By: Gavin Thorne, Gavin Thorne is an investigative journalist based in Washington, D.C., covering legislative affairs, special-interest influence, and the mechanics of congressional power.