Representative Anna Paulina Luna (R-FL) has found herself at the center of a controversy over the use of artificial intelligence in legislative drafting, after screenshots of an amendment summary for the 2027 National Defense Authorization Act appeared to show AI-generated text embedded in a congressional document. Luna initially acknowledged that her staff had used AI to correct draft text, but after the post drew sharp speculation that the technology was being used to write bills themselves, she issued a firm denial and a sharper clarification: the AI tool, identified as Claude, was used only for spellcheck and grammar review on a summary of the amendment, not on the actual bill language.
How the Screenshots Revealed the AI Trail
Accounts on X began circulating images of an amendment summary that contained a conspicuous artifact: the line “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 presence of a timestamp and the direct invocation of Claude — the large language model developed by Anthropic — strongly suggested that the text had been generated or heavily modified by an AI assistant. The formatting error, likely introduced when a staff member failed to strip the AI’s internal dialogue from the final document, made the usage visible to anyone reading the file.
The incident immediately raised questions about how deeply AI tools are being integrated into the legislative process, and whether the safeguards meant to keep machine-generated content out of official government texts are actually working.
Luna’s Initial Response and the Swift Edit
Luna’s first public response did little to dampen the speculation. She posted that “staff used AI to correct a draft text and didn’t edit,” adding that “not a shocker. Most staff use it. I have told them to make sure they are double checking and more thorough.” That framing — correcting a draft — left ambiguous whether the AI had touched the amendment’s legal language or the summary meant for public and congressional review. It was a distinction that mattered, and critics on X were quick to assume the worst.
Within a short window, Luna edited her post to sharpen the distinction. The revised statement reads: “Yeah my staff used AI to spell/grammar check the amendment SUMMARY, not the actual amendment text itself.” The edit moved the controversy from a potential breach of legislative drafting norms to a more mundane, if still notable, admission of AI-assisted administrative work.
House Legislative Council Rules and the “No AI on Bill Text” Claim
Luna followed up with a second post that directly addressed the core concern: “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 🤣.”
That claim is significant because it points to an existing institutional barrier. The House Legislative Council, which is responsible for producing the official text of all House bills and amendments, operates under a prohibition on using AI for drafting. If that rule is consistently enforced, it would indeed mean that no bill or amendment text originating from the House could have been written by a language model. The leaked text, Luna argues, came from a separate document — a summary layer produced by staff as a convenience for review — not from the legislative language that carries legal force.
What the Incident Reveals About AI in Government Workflows
Even if Luna’s explanation is accepted at face value, the incident exposes something broader: the use of AI tools like Claude for routine government document work is already happening, and the boundaries around that use are not well understood either by the public or by the staffers employing the tools. A spellcheck operation on a summary sounds innocuous, but the fact that the AI’s output leaked into a public-facing document shows that these tools are being used without robust guardrails around what gets inspected before release.
The controversy also highlights a gap in current policy. While the House Legislative Council may be barred from using AI for drafting, there is no equivalent prohibition on congressional staff using large language models to summarize, edit, or proofread documents that accompany legislation. That distinction is likely to receive more scrutiny as similar incidents surface.
For AI practitioners and observers, the takeaway is a familiar one: the technology is entering every professional domain, including the legislative branch, faster than the formal rules can adapt. The tools themselves — Claude, ChatGPT, and others — are powerful enough to generate convincing legal-sounding text, but they also leave artifacts that can erode trust in the integrity of the process. Any organization deploying these tools in high-stakes writing environments needs a clear policy on what the AI can touch, a technical workflow that strips metadata and internal prompts from final outputs, and a human review step that catches exactly the kind of error Luna’s staff missed.
Who Should Monitor This Story
This development matters for anyone working in government technology, compliance, or AI governance. It also matters for the broader professional audience that relies on AI-assisted writing tools: the line between helpful editing and substantive generation is thinner than many teams realize, and a single poorly filtered output can create a reputational crisis. Readers in the US, UK, Australia, and Canada should watch how Congress responds — whether it will clarify the rules around AI use by staff, strengthen enforcement of existing prohibitions, or introduce new transparency requirements for any AI-assisted document work. In the meantime, the practical lesson for any team using large language models for document preparation is to audit outputs for traces of the AI’s internal prompt structure before those documents ever see public light.
