This is a research prototype. The data and analyses are preliminary and not yet validated — we'd welcome your .
A US congressional staffer mistakenly copied and pasted Claude's raw output, including metadata and timestamps, directly into a public amendment summary for a major defense bill.
A staffer in Rep. Anna Paulina Luna's office reportedly used Claude to prepare a summary for an amendment to the FY2027 National Defense Authorization Act and accidentally pasted chatbot transcript residue, including a timestamp and "Claude responded:" marker, into the public House Rules Committee record. The summary was reportedly later corrected. Luna confirmed that her staff used Claude for the summary, not to draft the amendment itself.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
Users anthropomorphizing, trusting, or relying on AI systems, leading to emotional or material dependence and inappropriate relationships with or expectations of AI systems. Trust can be exploited by malicious actors (e.g., to harvest personal information or enable manipulation), or result in harm from inappropriate use of AI in critical situations (e.g., medical emergency). Overreliance on AI systems can compromise autonomy and weaken social ties.
Also classified as
Congress's internal rules on AI use are poorly understood, unenforced, and lack centralized oversight, leading to a 'free-for-all' environment.
The staffer over-relied on the AI system to draft the amendment summary and pasted the output directly into the public record without proper review, including the AI's metadata.
Human
Due to a decision or action made by humans
Unintentional
Due to an unexpected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed
The incident was caused by a human staffer's careless action of copying and pasting raw AI output into a public document, which was an unintentional oversight.
A US congressional staffer accidentally included Claude metadata in a public amendment summary for the National Defense Authorization Act. While the incident itself caused negligible direct harm, it highlights the broader, unregulated adoption of commercial AI tools by legislative staff, posing minor long-term risks to information security and government administrative governance.
The AI systems (Claude, ChatGPT, Copilot) are described as being used by staffers to augment their capabilities by drafting summaries, writing speeches, sorting mail, and researching policy, all of which require human oversight and decision-making.
Limited Risk: The AI systems used are general-purpose chatbots (Claude, ChatGPT, Gemini, Copilot). According to the definitions, chatbots and AI-generated content fall under Risk Level 3, which requires transparency obligations so users know they are interacting with AI.
The primary use of Claude in the incident was to spellcheck and summarize an amendment, acting as a writing assistant. Other mentioned uses include sorting mail and compiling databases, which align with content curation.
The reports provide clear, consistent, and detailed accounts of the specific copy-paste incident involving Rep. Luna's staff and Claude, as well as the broader context of AI adoption in Congress. The facts are well-documented with direct quotes from lawmakers and staffers.
The incident could be viewed purely as a minor human copy-paste error rather than an AI safety incident, but the systemic lack of AI governance in Congress makes it a relevant case study for AI safety and policy.
Technology
Chatbot Output Formatting
AI outputs include metadata and timestamps that can be copied accidentally.
No Automated Submission Checks
Public record submission systems do not screen for AI-generated text markers.
Human Factors
Negligent Copy-Pasting
Staffer copied chatbot text including 'Claude responded' into public record.
Overreliance on AI Chatbots
Staffers trust AI outputs without conducting thorough manual proofreading.
Ignorance of AI Guidelines
Staffers are unaware of the rules governing AI use on Capitol Hill.
Process and Methods
Absence of Peer Review
No secondary review was performed on the summary before public submission.
Informal Copy-Paste Workflows
Workflows rely on manual copy-pasting from external chatbot interfaces.
Management
Lack of Office Level Policies
Individual offices fail to implement and maintain written AI usage policies.
Inadequate Staff Training
Management does not train staffers on safe and compliant AI integration.
Prioritizing Speed Over Safety
Management encourages rapid AI use to save time without safety oversight.
Regulatory Environment
Unenforced AI Guidelines
Existing rules are poorly understood and there is no active enforcement.
Absence of Standardized Mandates
No uniform, binding regulations exist across all congressional offices.
AI system involved
Identified by automated extraction; can differ from the alleged developers above.
