US judges adopt AI for legal research despite mounting errors

US federal judges are increasingly experimenting with generative AI to help with legal research, case summaries, and routine orders, despite recent high-profile mistakes where AI-generated errors went undetected in court rulings. The trend highlights a growing tension between judicial efficiency and accountability, as judges face fewer consequences than lawyers when AI mistakes slip through—yet their errors carry the force of law.

The big picture: While lawyers have faced sanctions and embarrassment for submitting AI-generated briefs with fabricated cases, judges are now making similar mistakes with far greater consequences.

• In June, a Georgia appellate court judge issued an order relying on made-up cases that went uncaught.
• In July, a federal judge in New Jersey had to withdraw an opinion after lawyers complained it contained hallucinations.
• In August, a Mississippi federal judge reissued a civil rights decision riddled with errors but refused to explain what caused them, writing “No further explanation is warranted.”

What judges are using AI for: Early adopters like Federal Judge Xavier Rodriguez and Magistrate Judge Allison Goddard describe using AI tools for tasks they consider “routine” and low-risk.

• Rodriguez uses AI to summarize cases, identify key players, create timelines, and generate questions for attorneys ahead of hearings.
• Goddard keeps Claude “open all day” and uses it as a “thought partner” to understand technical issues and organize messy documents.
• Both encourage their clerks to use AI tools, with Goddard specifically recommending Claude because it doesn’t train on user conversations by default.

The accountability gap: Unlike lawyers who face scrutiny when AI mistakes surface, judges operate with less transparency and face fewer consequences.

• “When the judge makes a mistake, that’s the law,” warns Judge Scott Schlegel from Louisiana’s Fifth Circuit Court of Appeal.
• Attorneys can be sanctioned, have motions dismissed, or lose cases when opposing parties catch their AI errors.
• Judges cannot easily reverse themselves months later, and judicial mistakes in child custody or bail proceedings carry “pretty significant consequences.”

Drawing boundaries: Judges struggle to define which tasks are appropriate for AI assistance versus those requiring human judgment.

• Rodriguez considers case summarization and timeline creation “relatively risk free” but draws the line at using AI to predict bail eligibility.
• Research by Worcester Polytechnic Institute’s Erin Solovey found that “the line between what is appropriate for a human judge to do versus what is appropriate for AI tools to do changes from judge to judge and from one scenario to the next.”
• The Sedona Conference, an influential legal think tank, published guidelines in February outlining “safe” AI uses like legal research and document searching, while warning that “no known GenAI tools have fully resolved the hallucination problem.”

Why this matters: The stakes for judicial AI adoption are uniquely high because mistakes can erode public faith in court legitimacy.

• Nearly a quarter of federal civil cases involve at least one unrepresented party, increasing the likelihood of AI-generated errors going unchallenged.
• Some in the AI sector suggest that AI’s supposed objectivity could make it better than “fallible humans” at judging, potentially pressuring judges to adopt the technology.
• Judge Schlegel warns of a “crisis waiting to happen” that would dwarf the problem of lawyers submitting filings with fabricated cases.

What they’re saying: Judges express both enthusiasm and caution about AI integration in the legal system.

• “I think there’s been an overreaction by a lot of judges on these sanctions. The running joke I tell when I’m on the speaking circuit is that lawyers have been hallucinating well before AI,” Rodriguez said.
• “I’m not going to be the judge that cites hallucinated cases and orders. It’s really embarrassing, very professionally embarrassing,” Goddard emphasized.
• “If you’re making a decision on who gets the kids this weekend and somebody finds out you use Grok and you should have used Gemini or ChatGPT—you know, that’s not the justice system,” Schlegel concluded.