According to a new survey published on Monday, 60% of federal judges in the United States use at least one AI tool in their job.
Since the use of AI in legal work took off after OpenAI’s ChatGPT was released in 2022, Northwestern University researchers believe their study is the first random-sample investigation of AI usage among federal judges.
While 60% of judges surveyed said they used AI at least occasionally for tasks like document review, legal research, and document drafting and editing, only 22% said they used the technology daily or weekly, indicating that far fewer judges are using AI in their daily work, and 38% said they had never used AI in their roles as judges.
Daniel Linna, a professor of law at Northwestern, co-authored the paper with V.S. Subrahmanian, a professor of computer science at Northwestern, U.S. District Judge Xavier Rodriguez, and Siyu Tao, a student at Northwestern. “Judges see the potential…here,” Linna said. “There are risks, of course, but if we do this correctly, there are benefits.”
A growing number of attorneys have been sanctioned by the court for submitting papers that cited cases that were “hallucinated” by AI, and two judges acknowledged last year that staff workers had used AI to help produce recent court orders that had errors.
From a random sample of 502 federal bankruptcy, magistrate, district court, and appellate court judges, 112 federal judges provided replies for the new study. According to Linna, that’s a significant enough response to make broad inferences about how federal judges are handling AI.
In addition to AI tools created especially for legal use, such as CoCounsel, Westlaw AI-Assisted, and Deep Research (all owned by Thomson Reuters, the parent company of Reuters), Protégé or Lexis+ AI, Vincent AI, Harvey, and Legora, the survey asked judges about their use of well-known large language models, such as ChatGPT and Claude.
The researchers discovered that judges were more likely to employ legal-specific AI tools than general ones.
At 30%, the respondents said they used AI more for legal research than for any other work. At over 16%, reviewing documents was the next most popular use.
Of the judges who responded, one in three stated that they either support or allow the use of AI in their chambers, while 20% officially forbid it. While 24% stated they have no explicit policy, nearly 18% discourage the usage of AI without outright prohibiting it.
According to Linna, over 45% of judges in the study stated that court administration did not offer AI training. This data can assist guide policy decisions on court spending and training.
