Rep. Anna Paulina Luna (R-FL) is denying that her staff used AI to draft an amendment to the 2027 National Defense Authorization Act, after a screenshot circulating on X showed an apparent Claude prompt embedded directly in the amendment's summary text. The screenshot, drawn from an amendment summary, included the visible string "Claude responded:" followed by language describing operations at the southwest land border. Luna says staff used the Anthropic chatbot for spellcheck on the summary, not for the bill text itself.
The visible artifact in the screenshot read, in part: "Identical to H.R. 100 (118th Congress).11:25 AM????Claude responded: Requires the Secretary of Defense to designate Department of Defense activities, support, and operations at the southwest land border as a named operation with…" The leftover prompt-and-response framing is the kind of artifact that typically appears when someone pastes chatbot output without cleaning it up.
Luna's first response on X did little to settle the matter. She wrote that "staff used AI to correct a draft text and didn't edit," wording that read to many as a concession that Claude had touched the amendment text itself.
“staff used AI to correct a draft text and didn't edit.”— Anna Paulina Luna, U.S. Representative (R-FL)
Key facts
- 01Rep. Anna Paulina Luna (R-FL) denies AI wrote the text of her amendment to the 2027 National Defense Authorization Act.
- 02A screenshot circulating on X showed a Claude prompt embedded in the amendment summary, including the line 'Claude responded:'.
- 03Luna says staff used Anthropic's Claude only to spell- and grammar-check the summary, not the bill text.
- 04She says all House bill text comes from the House Legislative Council, which is prohibited from using AI.
- 05Arizona state representative Alexander Kolodin has previously said he uses ChatGPT to draft state legislation.
She then edited the post to draw a sharper line between the summary and the legislative text. Luna also added a follow-up post: "FYI NO Legislation is ever drafted with AI. All bill text from the House comes from the House Legislative Council which is prohibited from using AI. The screenshot you're referencing is an AI summary of the bill that's also used for spellcheck, cmon man 🤣"
The House Legislative Council is the nonpartisan office that drafts statutory text for House members. Luna's account places responsibility for the actual amendment language there, with Claude confined to a polishing role on the summary that accompanies it.
“Yeah my staff used AI to spell/grammar check the amendment SUMMARY, not the actual amendment text itself”— Anna Paulina Luna, U.S. Representative (R-FL)
Her broader point — that congressional staff routinely use AI tools — is harder to dispute.
Luna acknowledged as much in a separate post, saying she had instructed her team to double-check and be more thorough.
“Not a shocker. Most staff use it. I have told them to make sure they are double checking and more thorough.”— Anna Paulina Luna, U.S. Representative (R-FL)
The episode is part of a widening pattern of AI tools surfacing in official documents where they were not disclosed. Federal judges have repeatedly caught lawyers filing briefs with fabricated citations generated by chatbots. City officials in Brazil approved an ordinance that was later revealed to have been written with ChatGPT, with the council members unaware. Arizona state representative Alexander Kolodin has said openly that he uses ChatGPT to help draft state-level legislation.
What sets the Luna incident apart is the artifact itself. A pasted "Claude responded:" string is the AI-era equivalent of leaving track-changes on — it documents the tool's involvement in the document at the moment of publication, regardless of what role staff intended it to play. The amendment summary is a public-facing document that members of Congress and their staffs rely on to understand what they are voting on.
Anthropic markets Claude heavily to enterprise and government users, and the federal government is an increasingly important customer for every frontier lab. A high-visibility incident in which Claude output ends up pasted verbatim into a congressional document — even a summary — is the kind of thing model providers prefer to avoid, because it invites questions about deployment hygiene rather than model capability.
The unresolved question is how common this already is across Capitol Hill. Luna's own framing — that most staff use AI tools — suggests the answer is "widely," with no consistent disclosure standard. The House Legislative Council's prohibition covers the office that drafts statutory text, but it does not cover the member offices that draft summaries, talking points, press releases, and constituent correspondence around that text.
Expect this to accelerate the push for explicit AI-use disclosure rules in congressional offices, and for procurement-side conversations about which models are cleared for which legislative workflows. The market opportunity for Anthropic, OpenAI, and others in federal staff work is real and growing, but it runs through governance frameworks that barely exist yet. Stories like this one are how those frameworks get written — reactively, one screenshot at a time.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
