A Congressional Staffer Copy-Pasted a Chatbot Answer Straight Into the National Defense Authorization Act

A summary attached to one of Rep. Anna Paulina Luna's amendments to this year's National Defense Authorization Act contained an opening line that made no sense to anyone reading it. It was a chatbot's answer, timestamp and all, pasted directly into the public record.

The screenshot spread on social media. Luna, a Florida Republican, wasn't embarrassed. "Not a shocker," she wrote. "Most staff use it."

She later told The Washington Post the mistake was procedural, not substantive. "A lot of staff will use ChatGPT, Claude, Grok. I personally like Grok better, but my staffer prefers Claude," she said. "I'm not gonna knock him for that. There's nothing illegal about him doing a summary."

She's right about the legality. Nothing in House rules bars a staffer from using AI to draft a summary of legislative text. Her staffer's actual error was leaving the chatbot's fingerprints visible for the public to see.

Widespread, Fast, Barely Governed

According to The Washington Post, AI use inside the Capitol has moved well past isolated incidents. Lawmakers and staffers are using chatbots to write speeches, draft news releases, sort constituent mail, prepare questions for committee hearings, and write amendments. This work traditionally fell to aides, speechwriters and policy staff.

Both the House and Senate have formally cleared members and staff to use commercial AI tools including Microsoft Copilot, ChatGPT, Google's Gemini, and, in the House, Anthropic's Claude, with additional tools under review. The House bought 6,000 Copilot licenses last year and actively encouraged staff to use them, the Post reported.

That's an institution-wide bet on AI adoption happening faster than the institution can write rules for it. The Post's reporting, based on interviews with more than two dozen lawmakers and staffers plus internal House and Senate policy documents, found the guardrails that exist are thin, unevenly applied, and rarely enforced. It's not even clear how many individual congressional offices have written AI policies of their own. That's been left to each office to decide.

Most strikingly: nobody The Post talked to could point to a single instance of a staffer being formally disciplined for violating whatever AI rules do exist.

The Conflict of Interest

The same Congress writing chatbot-assisted speeches about how to regulate AI is using products built by the tech companies it's supposed to be regulating, including OpenAI, Google, Microsoft and Anthropic. That's not illegal, and it's not necessarily corrupt. But it's a real conflict of interest that deserves scrutiny, and Congress hasn't offered much explanation for how it manages that tension internally.

There's a reasonable defense here too. Ghostwriting isn't new on Capitol Hill. Staffers have drafted speeches for their bosses for as long as Congress has existed, and nobody calls that scandalous. A chatbot doing a first-pass summary of amendment language isn't fundamentally different from an intern doing the same job, faster. Luna's underlying argument, that having AI draft a summary is no different from having a staffer draft one, isn't unreasonable on its face.

But a staffer can be fired, cross-examined, or held accountable for a factual error. A chatbot can't testify to how it reached a conclusion, and it has no stake in getting classified defense provisions or budget numbers exactly right. When AI tools are drafting language for something as consequential as the NDAA, the annual defense policy bill, the difference between "a staffer wrote it" and "a chatbot wrote it and nobody caught the artifact before it hit the public record" starts to matter.

Enforcement and Next Steps

Neither of the source reports names a single congressional office with a comprehensive, enforced AI policy. Sen. Tommy Tuberville's office is mentioned in connection with AI use, though the specifics of what rules, if any, apply to his staff aren't spelled out in available reporting.

Open questions remain: How many of the hundreds of amendments filed to bills like the NDAA this year contain AI-generated content nobody flagged? What happens when a chatbot introduces a factual error into legislative text that isn't as obvious as a leftover timestamp? And with no known disciplinary cases on record in either chamber, is there any actual enforcement mechanism waiting to be used, or is the rule effectively unenforceable by design?

Congress hasn't answered those questions. It has, however, already answered the practical one. Whether to use the tools at all was decided months ago, office by office, with no public vote and no unified rulebook.