"I accidentally deleted my original brief because I was overwhelmed, so I used AI to try and replicate my prior work."
That's what Assistant U.S. Attorney Rudy Renfer reportedly told a federal magistrate judge today during a show-cause hearing in Fivehouse v. U.S. Department of Defense. It's a remarkable admission from an attorney who, until now, had offered a much vaguer explanation for why his court filing was stuffed with fabricated legal citations.
What Renfer Actually Filed
Magistrate Judge Robert Numbers identified at least six categories of problems in Renfer's brief. The filing included fabricated quotations from multiple circuit court opinions, including Ohio Valley Environmental Coalition v. Aracoma Coal Co. and Dow AgroSciences v. National Marine Fisheries Service. It also contained two fabricated quotes from the Code of Federal Regulations. These weren't minor paraphrasing errors. The quotes and case holdings were simply made up.
The person who caught all of this wasn't opposing counsel from a major law firm. It was the plaintiff himself, Derence Fivehouse, a retired Air Force colonel and practicing attorney who filed the lawsuit pro se (meaning without a lawyer) to challenge a DOD policy restricting GLP-1 weight loss medications for TRICARE for Life participants.
Before today's hearing, Renfer had blamed the errors on the "inadvertent filing of an unfinalized draft document." The U.S. Attorney's office in Raleigh, North Carolina refused to answer questions about whether AI tools were involved. Judge Numbers wasn't satisfied. In a March 2 order, he demanded that senior leaders from the entire U.S. Attorney's Office for the Eastern District of North Carolina appear to explain why Renfer shouldn't be sanctioned and why the office shouldn't be held jointly responsible.
The AI Excuse Has a Familiar Shape
Renfer's admission fits a pattern that's become distressingly common in federal courts. The fabricated citations he filed look exactly like AI hallucinations (instances where a chatbot generates plausible-sounding but completely fictional legal citations). In February 2026 alone, a Kansas federal judge fined five lawyers for submitting AI-generated fake citations, and an Am Law 100 firm was accused of the same thing for the second time.
The "I deleted my work and used AI to recreate it" explanation is new, though. It raises an obvious question: if Renfer had already written a legitimate brief with real citations, why would an AI tool generating the same arguments produce fake ones? The answer is that AI chatbots don't retrieve information from legal databases. They predict what text should come next based on patterns. They generate citations that look right but may not exist.
Renfer has been a bar member for nearly 30 years and has worked at the U.S. Attorney's office since 2009. This isn't a junior associate who didn't know better.
A Warning for Anyone Using AI to Draft Professional Documents
The potential consequences here range from monetary fines to contempt proceedings to suspension from practicing before the court. For a 30-year DOJ veteran, that's a career-defining moment.
The lesson is one that keeps getting repeated because people keep not learning it: AI writing tools can produce text that reads like professional work but contains fabricated facts. This is true for legal citations, medical references, financial data, and technical specifications. If your name goes on the document, you're responsible for every claim in it, regardless of what tool generated the first draft.
Every major AI provider, including OpenAI and Anthropic, warns in their documentation that their models can produce inaccurate information. The tools aren't broken. They're being used for something they were never designed to do: serve as a source of factual record.