Rep. Anna Paulina Luna Uses AI in Defense Amendment

Multiple outlets report that an amendment description for the 2027 National Defense Authorization Act posted on the House Rules Committee website contained an embedded artifact reading "11:25 AM????Claude responded," language that several reporters and The New Republic identified as an output trace from Anthropic's assistant Claude (The New Republic; Gizmodo). The Rules Committee text was later revised, according to The New Republic. Screenshots of the original text circulated on X and triggered coverage from outlets including The Verge, Gizmodo, Mediaite, and others.

Rep. Anna Paulina Luna posted on X, "Yeah my staff used AI to spell/grammar check the amendment SUMMARY, not the actual amendment text itself," a reply documented by The Verge, Gizmodo, and Mediaite. The Verge reports Luna also wrote, "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." Mediaite recorded a reporter exchange in which Luna defended her staff's use of AI for spell-checking.

Artifacts such as timestamps and assistant-attributed lines commonly appear when users copy-and-paste conversation logs or model outputs without sanitizing system messages. Industry practitioners and documentation for chat assistants note that conversation transcripts may include internal markers like "assistant responded" or metadata when exported or copied; public reporting indicates the visible phrase matched Claude-style output (Gizmodo; The New Republic). In comparable operational settings, teams that paste model-generated summaries into downstream documents risk leaving identifying traces that reveal provenance and reduce document hygiene.

Reporting highlights a governance and compliance tension between informal staff workflows and formal legislative drafting processes. The Verge documents Luna's claim that the House Legislative Counsel does not use AI for official bill text, a procedural point that intersects with broader debates over permissible AI use in government drafting. Industry observers have previously flagged the need for clear rules and audit trails when model outputs are used in decision-making or formal documents.

Observers should track whether congressional offices or House committees issue formal guidance on the permissible scope of AI assistance for staff work, including disclosure requirements and auditability of sources. Watch for follow-up reporting or statements from the House Rules Committee, the House Legislative Counsel, or Luna's office clarifying whether the amended public text originated with House counsel and how summaries are prepared. Also watch for any public protocols from vendors or platform providers about removing metadata or system messages before external use.

Documentation and repeatable sanitation steps for model outputs are practical controls to avoid leaking provenance artifacts into public records. Industry teams that integrate assistant outputs into formal documents generally adopt explicit provenance tagging, redaction, and verification steps; those patterns are likely to be relevant to public-sector offices that may also use assistants for drafting or summarization tasks.