Washington is hooked. And it's not on the usual K Street stuff. It's on ChatGPT.
New data from a government transparency tracker shows at least 70 House offices used identifiable AI tools in the first half of 2026. That's one in six congressional districts. And that's just what we can see. The real number is almost certainly higher, because many offices use free versions or generic tools that don't show up in payment records.
The numbers come from LegiStorm, a nonpartisan tracker that scours congressional financial disclosures. They found Democrats leading visible spending on AI, with at least 45 offices using tools like ChatGPT Plus, Claude Pro, and other paid AI services. Republicans aren't far behind, with 25 offices. But the gap tells a story.
Democrats, who have been pushing for AI regulation, are also the ones spending the most on the very technology they want to control. That's not hypocrisy. That's called learning the terrain. But it does raise questions.
The AI Arms Race in Congress
Let's be clear: this isn't about lawmakers writing their own speeches with AI. Well, maybe some of them. But the broader picture is that congressional staffers are drowning in work. They draft bills, respond to constituents, prep for hearings. AI is a lifeline.
One senior staffer told me, off the record, that his office uses ChatGPT to summarize legislation and draft emails. "It saves us hours a day," he said. "We'd be idiots not to use it."
And he's right. But here's the catch: the same technology that helps a staffer draft a constituent response can also be used to draft a press release that spins the truth. Or to generate fake statements from opponents. The line between tool and weapon is thin.
"It saves us hours a day," one staffer said. "We'd be idiots not to use it."
The spending is small potatoes — a few hundred bucks a month per office. But the implications are huge. Congress is now using the same AI that it's trying to regulate. That creates a conflict of interest that nobody wants to talk about.
Regulating What You Use
Lawmakers have introduced dozens of AI bills. Some would require transparency when AI is used in political ads. Others would create a new agency to oversee AI safety. But how do you regulate an industry you're already embedded in?
"It's like asking a smoker to write the surgeon general's warning," says Sarah Miller, a tech policy analyst at the R Street Institute. "They know it's bad for them, but they can't quit."
Miller's point is sharp. Congress has a tiger by the tail. They want to set rules for AI, but they're also enjoying the benefits. And the benefits are real. Faster constituent services. Better bill drafting. More efficient research.
But here's what keeps me up at night: AI in Congress is not just a tool. It's a power multiplier. The offices that use AI well will be more productive. They'll answer more constituent emails, draft more bills, and raise more money. The offices that don't? They'll fall behind. That's not a level playing field.
The Undercount Problem
The 70 offices using identifiable AI tools is a floor, not a ceiling. Many offices use free versions of ChatGPT or other tools that don't require a credit card. Others use AI through platforms like Microsoft Copilot, which is bundled into Office 365 and doesn't show up as a separate expense.
So the real number is probably 200 or more offices. That's nearly half of the House. And that's just the House. The Senate hasn't been tracked as closely, but you can bet they're not immune.
This undercount is a problem for anyone who wants to oversee AI use in government. You can't regulate what you can't see. And right now, we're flying blind.
What This Means for Democracy
Here's the uncomfortable truth: AI is already shaping the work of Congress. And it's doing it without any public debate. No hearings were held before staffers started using ChatGPT. No ethics rules were updated. It just happened.
That's not necessarily bad. AI could make Congress more efficient. Maybe it'll help them write better laws. But it could also make them lazy. Why read a 500-page bill when ChatGPT can summarize it in 30 seconds? The problem is, summaries lose nuance. And nuance is where the devil lives.
Take a recent example. A House committee used AI to analyze public comments on a proposed rule. The AI flagged certain themes and ignored others. The final rule reflected those themes. Nobody noticed until a watchdog group did its own analysis. The group found that the AI had systematically downplayed comments from consumer advocates. Coincidence? Or bias in the algorithm?
That's the kind of thing that happens when you outsource your thinking to a machine. The machine doesn't have your values. It has data.
So what do we do? We could ban AI in Congress. But that's like banning the internet. It's not going to happen. Instead, we need radical transparency. Every AI tool used by a congressional office should be disclosed. Every AI-generated output that's used in official documents should be labeled. And there should be an independent watchdog to audit AI use.
Is that too much to ask? Maybe. But consider the alternative. In a few years, you won't know if your congressman's speech was written by a human or a machine. And if you can't tell, can you trust it?
This is not a hypothetical. It's happening right now. And the first step to fixing a problem is admitting it exists. Congress needs to admit that it's using AI. And then it needs to decide what that means.
I'll be watching. So should you.
