U.S. Rep. Anna Paulina Luna, R-Fla., recently sparked controversy when her office used Anthropic's Claude to work on a defense bill amendment brief. After the incident came to light, she first admitted that staff had used AI to "proofread" or "correct" it, and then emphasized that AI was only used for summaries and spelling grammar checks, and was not involved in drafting the main text of the bill.
The controversy started when a screenshot was circulated on social platforms. The text in the screenshot showed words related to Claude. Based on this, the outside world suspected that her team directly used AI to write the content of the defense bill amendment. Luna later responded by writing that the staff used AI only to check the spelling and grammar of the summary of the amendment, not for the text of the amendment itself; she also said that the official bill text of the House of Representatives was drafted by the Congressional Legal Advisory Agency, which is prohibited from using AI.
In her initial response, Luna's statement was once interpreted as admitting that staff had used AI to "correct the draft text," further fueling doubts. She later revised her statement to say that AI was only used to "spellcheck the amendment SUMMARY" and repeated that "no legislation will ever be written by AI." This inconsistent statement made her the focus of public opinion, and also brought the issue of "whether AI is used for substantive legislative work" to the forefront again.
Reports show that the content being discussed is related to the 2027 National Defense Authorization Act. The core of the dispute is not just whether Claude is used, but the depth of AI involvement. Luna’s current statement is that AI only participated in text polishing and spelling checking at the summary level, and did not participate in the writing of the formal bill text. But as screenshots and her multiple public responses continue to circulate online, the incident has evolved into a public debate about the boundaries of government offices, the legislative process, and generative AI.