Rep. Luna denies Claude wrote an amendment to the US defense budget

Official documents for an amendment to the US defense budget contained a leftover fragment of a conversation with Claude. Rep. Anna Paulina Luna says her staff used AI only to check the text, not to draft legislation.

What showed up in the screenshot

Several X users spotted something unexpected in the summary of an amendment to the National Defense Authorization Act for 2027 (NDAA), the main annual US military budget. In the middle of the official text sat the line: "11:25 AM ???? Claude responded:". It was followed by a brief rundown of the amendment's requirements, which would give the Department of Defense's operations on the southwest border the status of an official named operation.

The screenshot spread across X instantly. The conclusion seemed obvious: someone on the congresswoman's staff had copied a reply from a Claude chat straight into an official document without deleting the conversation header. The criticism came immediately.

How Luna responded

The Florida Republican responded quickly. She said Claude was used solely to "spell-check" the amendment's summary, not the legislative text itself.

"No bill is ever drafted with AI," Luna wrote.

Her initial response, however, was ambiguous and gave the impression that AI had not been used at all. Luna had to clarify her position separately: the AI use involved only the supporting summary, not the text of the amendment.

The distinction matters. An amendment summary is an explanatory document read by journalists, lobbyists, and fellow lawmakers, and that is where the trace of Claude turned up. According to the congresswoman, the amendment text itself was written only by people.

A big problem behind a small slip

The incident exposes a systemic problem: the US Congress still has no unified policy on using AI in legislative work. What happened in Luna's office is a symptom, not an exception.

• There is no common standard for using AI tools on bills
• Some offices have introduced their own bans on external chatbots for drafting
• Amendment summaries end up in public documents and shape how legislative provisions are perceived
• A conversation artifact in an official document points to weak oversight of workflows
• The NDAA draws a huge number of amendments every year, putting heavy pressure on staff

Tight deadlines push staffers to use whatever tools are available, AI included, without clear protocols and without realizing that "raw" output can end up in a public document.

What it means

The incident in Luna's office is less a political scandal than a reminder of the operational risks of unstructured AI use. Without internal rules, a line like "Claude responded:" can easily land in an official document.

Congress will have to settle on a unified position: a ban, mandatory disclosure, or rules for acceptable use. Until that happens, incidents like this will keep recurring.