The Pulse
U.S. Judiciary Tells Courts Not to Delegate Decisions to AI
The Judicial Conference says federal courts must not hand core judicial functions, including case adjudication, to artificial intelligence. Interim guidance also makes judges and court staff accountable for work produced with AI assistance.
AI.info Team ·
Federal courts draw a line around judicial judgment
The U.S. Judiciary is warning federal courts not to hand core judicial functions to artificial intelligence, including decisions and case adjudication. The guidance came as the Judicial Conference received an update on the branch’s work to establish policies for AI use.
Judge Robert J. Conrad Jr., director of the Administrative Office of the U.S. Courts and secretary of the Judicial Conference, said the branch has issued interim guidance while it studies the questions raised by AI. “We are moving forward with a sense of urgency and focus,” Conrad said in remarks reported by the U.S. Courts.
The guidance does not prohibit every use of AI in court operations. It establishes human responsibility as the governing rule: Judiciary users must review work created with AI assistance, independently verify its content, and remain accountable for the final product.
More than 60 questions are under review
Conrad appointed an advisory AI task force in 2025 to coordinate the Judiciary’s response to the technology. The group has identified more than 60 distinct issues and divided its work among seven subject-matter subgroups, according to the Judicial Conference update.
The task force is examining how AI could affect the work of judges, court staff and the wider federal court system. The current guidance is temporary, with additional policies expected as the branch resolves the questions under study.
The limits are clearest where AI would replace judicial judgment. Courts have been cautioned against delegating decision-making or case adjudication to an AI system. The Judiciary also has urged extreme caution when AI is used to help address novel legal questions, according to the branch’s 2025 annual report.
Human review stays with the person using the system
The guidance places responsibility on the individual who uses or approves an AI tool. Court personnel must check generated material rather than treat a system’s output as authoritative, and they may need to consider whether AI use should be disclosed.
The annual report says disclosure decisions should account for ethical obligations, local rules and practices, confidentiality, professional duties and the integrity of the Judiciary. Those considerations matter in court settings because generated text can enter filings, internal work products or records that affect litigants and public access to proceedings.
The policy approach gives courts room to test AI for administrative and research tasks while reserving adjudication for judges. It also avoids treating an AI-generated result as an independent source of legal authority: responsibility remains with the human official who relies on it.
The AI policy arrives alongside a broader technology overhaul
The Conference received the AI update alongside progress reports on two other Judiciary initiatives. The first is a redesign of the Case Management/Electronic Case Files system, which the branch says has become outdated and vulnerable to security breaches.
The first component of the replacement system is scheduled for release before the end of 2026. Conrad said all new district court cases are expected to move into the new case-management system by the end of 2027, followed by appellate and bankruptcy courts.
The Conference also reviewed the Judiciary’s request for authority to directly manage mission-essential properties now handled by the General Services Administration. A bipartisan group of senators introduced legislation on July 30 to create a pilot covering selected properties in no more than 10 judicial districts.
What the warning means for court users
The Judiciary’s language stops short of treating AI as forbidden technology. It instead draws a boundary between assistance and authority: software may help with work, but it cannot assume the responsibilities assigned to judges and court officials.
That distinction will shape future guidance as the task force works through its remaining questions. For now, the operational rule is direct: an AI system may assist Judiciary work, but a human user must verify the result and answer for it, and no court may delegate the decision itself.