Claude drafted a defense bill amendment — or did it?
rep. anna paulina luna admitted staff used anthropic's claude ai to spellcheck an amendment summary for the 2027 defense authorization act, denying ai wrote the legislation itself. the incident highlights the increasing invisible integration of llms into government drafting processes.
a screenshot shared on x revealed an amendment summary for the 2027 national defense authorization act containing the phrase 'claude responded,' exposing the use of anthropic's ai in a congressional office. rep. luna initially stated staff used ai to correct draft text without editing, later clarifying it was used only for spellchecking the summary and insisting the council drafting the actual bill text is prohibited from using ai. the incident fuels existing concerns over the unvetted integration of generative ai in sensitive legal and political workflows, following past cases of lawyers submitting fake citations generated by chatbots and brazilian officials passing chatgpt-written ordinances. as llms become ubiquitous workplace tools, the line between human and machine-generated legislative text blurs, raising questions about oversight and accountability.