More Than 60% of Federal Judges Have Used AI — But Consistent Adoption Remains Rare

A first-of-its-kind random-sample survey of federal judges reveals what many AI adoption studies obscure: the difference between having tried something and actually relying on it. Generative AI has reached the federal judiciary, but daily use is the exception, not the rule — and the data exposes a pattern that mirrors AI rollouts across most professional institutions.

Key Takeaways

• Majority exposure, minority habit: 61.6% of surveyed federal judges have used at least one AI tool in their judicial work — but only 5.4% use AI daily and 17% weekly. The largest group uses AI monthly or rarely, suggesting adoption is exploratory rather than embedded.
• Legal-specific tools dominate over general-purpose AI: Judges strongly prefer AI integrated into established research platforms (Westlaw AI leads at 38.4%) over general-purpose tools like ChatGPT (28.6%). Familiarity, perceived reliability, and institutional trust drive tool selection more than raw capability.
• Training gaps are limiting adoption — and demand is unmet: Nearly half of judges (45.5%) received no AI training from court administration. Yet among those who did receive training, 73.8% attended — suggesting the barrier isn't resistance to AI, but lack of structured access to learn it.

The survey, conducted by Northwestern University researchers in collaboration with the New York City Bar Association, found judges nearly evenly split between optimism (43%) and concern (42%) about AI's role in the judiciary. Concerns focused primarily on hallucinations, skill atrophy among clerks, and the downstream effects on a generation of lawyers who will enter the profession dependent on AI for research and writing.

For enterprise AI leaders, the judiciary data offers a useful mirror: professional institutions adopt AI more slowly than consumer tools — but once structured training and trusted integrations are in place, uptake follows. The bottleneck is rarely attitude. It's access, governance, and fit.

Read the full article on LawSites