US Congressional Staffers Caught Copy-Pasting AI Responses into Official Legislative Records
US congressional staff members, including those working for Representative Anna Paulina Luna, were found using AI tools like Anthropic's Claude to draft legislative documents, resulting in copy-paste errors being published in the official record of the National Defense Authorization Act. The leaked text explicitly included the phrase "Claude responded," exposing the unregulated use of AI on Capitol Hill. This incident highlights a stark irony where the very lawmakers tasked with regulating artificial intelligence are relying on these tools without formal oversight or guidelines. It underscores the rapid, unchecked adoption of large language models in sensitive government operations, raising concerns about accuracy and accountability in policy-making. The copy-paste error occurred in the public record for the National Defense Authorization Act, where text containing "11:25 AM????Claude responded:" was left in the document. While some offices have individual rules against drafting documents from scratch using AI, Capitol Hill currently lacks comprehensive, enforceable regulations regarding AI usage by staff.
## BACKGROUND
Large language models (LLMs) like OpenAI's ChatGPT and Anthropic's Claude are increasingly used across various industries to summarize information and draft text. However, these models are prone to generating errors or hallucinations, and their use in government drafting raises unique security and ethical questions, especially given the lack of standardized federal guidelines for legislative staff.