A DOJ Attorney Delegated to AI After Losing a Draft, and the Brief He Filed Contained Fabricated Quotations
What happened
In Fivehouse v. U.S. Department of Defense, Assistant U.S. Attorney Rudy Renfer filed a response brief on behalf of the government that the court reportedly found contained fabricated quotations and misstatements of case holdings. The filing was not a rough draft or a document flagged before submission. It went to the court as the official position of the United States Department of Justice.
Renfer later admitted he had used AI to help draft the brief after losing an earlier version he had prepared. The sequence matters: the brief was not the product of AI from the start, but of a human who reached for AI as a reconstruction tool after a failure in his own workflow and did not verify what came back before filing it. The court reportedly noted that the submitted brief cited cases for propositions they did not stand for and included quotations that did not appear in the cited sources.
The problem did not stop at a single filing. The court reportedly identified similar fabricated quotations across other briefs signed by Renfer, which suggests the filing in Fivehouse was not an isolated lapse but part of a pattern that had gone undetected until the court examined the citations closely enough to trace them.
The stakes here are higher than in a private practice context. A government attorney in a federal case represents institutional authority. Fabricated citations submitted to a federal court create a record that opposing counsel, judges, and clerks must work against under the assumption that the cited authority is real. Every hour spent tracing a citation that leads nowhere is time compensating for a failure the filing attorney had an obligation to catch. The opposing party in Fivehouse was litigating against the Department of Defense and deserved adversarial submissions grounded in actual law.
The gap this incident points to is not simply a question of professional discipline. There is currently no requirement that an attorney disclose which portions of a brief an AI system generated, no standard for documenting what prompts produced what output, and no audit trail connecting the filed document to the tools used to produce it. A provable record of what a system contributed, at which step, and what the attorney actually reviewed before signing would not prevent every hallucination, but it would make the oversight failure visible, attributable, and correctable before the next filing.
Reported impact
- Affected parties
- Not publicly disclosed
- Harm type
- Not publicly disclosed
- Scale
- Not publicly disclosed
- Financial impact
- Not publicly disclosed
- Regulatory action
- Not publicly disclosed
Classification
Relevant governance controls
Governance control mapping is not available for this record.
- No controls mapped
Not publicly disclosed
Control mapping is analytical. It does not state that any control would have prevented the incident.
Sources and evidence
This record was researched and written by the Index. The event is also catalogued in the following database, which is listed for cross-reference.