More than 60% of federal judges responding to a new survey say they've used at least one AI tool in their judicial work. But dig into the numbers and the adoption is thinner than that headline suggests.
The study, published by The Sedona Conference with Northwestern University researchers, surveyed 502 randomly selected federal judges in December 2025. Of the 112 who responded, only 22.4% use AI tools on a weekly or daily basis. About 38% have never used any AI tool in their work at all.
What Judges Actually Do With AI
Legal research is the top use case. 30% of judges use AI for it, and 39.8% of their chamber staff do too. Document review comes in second at 15.5% for judges, 16.7% for staff.
The study tracked both general-purpose tools like ChatGPT, Claude, Copilot, and Perplexity alongside specialized legal AI products like CoCounsel (Thomson Reuters), Lexis+ AI (LexisNexis), and Harvey. Judges were more likely to reach for the law-specific tools than general-purpose chatbots, which makes sense - when you're making decisions that affect people's lives, you want something built for legal reasoning, not a general chat model.
The Training Gap
Here's the part that should concern legal tech companies and court administrators: 45.5% of judges said their court offered no AI training. Another 15.7% weren't even sure whether training existed. Meanwhile, only 25.9% of judges formally permit AI use in their chambers, and about 20% formally prohibit it. Nearly a quarter have no policy at all.
"AI is here; it's not going anywhere," said Daniel Linna, who led the research as Director of Law and Technology Initiatives at Northwestern Pritzker Law. "We need training, best practices and clear policies on how the technology is implemented."
One finding connects personal and professional adoption directly: judges who use AI in their personal lives are significantly more likely to use it professionally. That pattern mirrors what we see across every industry - the people who play with ChatGPT at home are the ones pushing for it at work.
The study is the first based on a random sample of federal judges rather than self-selected respondents, which makes the data more reliable than previous surveys. With 112 responses out of 502 surveyed, the response rate is modest but typical for judicial surveys. The judges were nearly evenly split between optimism and concern about AI's impact on the courts.