US lawmaker denies AI used to draft defense amendment following Claude screenshot controversy

US Representative Anna Paulina Luna, a Republican from Florida, has denied that her office used artificial intelligence to draft legislative text, following the circulation of screenshots that appeared to show references to Anthropic’s Claude chatbot. The controversy centres on an amendment summary for the 2027 National Defense Authorization Act, which included a timestamp and a response from the AI model outlining requirements for the Secretary of Defense.

Screenshots shared on the platform X displayed an amendment summary containing the text: "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 the chatbot reference in the document summary sparked immediate scrutiny regarding the use of generative tools in the legislative process.

Luna initially responded to the screenshots by stating that her staff had used AI to "correct a draft text," noting that she had instructed them to ensure they were "double checking and more thorough." This initial comment led to speculation on social media that her office was utilising the technology to write the actual bill text, a practice that carries significant regulatory and procedural implications.

Following the backlash, Luna edited her post to clarify the extent of the technology’s use. She stated that her staff used AI solely to "spell/grammar check the amendment SUMMARY, not the actual amendment text itself." She further explained that the screenshot referenced was an "AI summary of the bill that’s also used for spellcheck," dismissing the broader allegations with a laughing emoji.

Luna asserted that "NO Legislation is ever drafted with AI" in her office. She highlighted that all bill text in the House originates from the House Legislative Council, the body responsible for producing legislative language, which is explicitly prohibited from using artificial intelligence. Her comments underscore the strict procedural boundaries that remain in place for formal legislative drafting, despite the increasing integration of AI tools in other workplace environments.

The incident reflects a broader global trend of AI adoption in legal and political spheres, where the line between assistance and authorship is increasingly scrutinised. Previous instances have seen judges identify lawyers using AI chatbots to draft filings with fabricated citations, while other lawmakers, such as an Arizona state representative, have openly admitted to using ChatGPT for state-level legislation. In Brazil, city officials even unknowingly approved an ordinance written entirely by the technology.

Despite these widespread adoptions, Luna’s clarification reinforces the distinction between administrative assistance and legislative authorship. The House Legislative Council’s prohibition on AI usage remains a key safeguard in the US legislative process, ensuring that the formal text of bills is produced without generative input, even as staff members utilise these tools for preliminary editing and summarisation tasks.