A large-scale trial in Pakistan found that a custom AI tool for judges, combined with training, boosted resolved cases by 6.3 percent with no obvious drop in judgment quality. With a backlog of 2.26 million cases and fewer than two judges per 100,000 people—compared to 22 in the EU and eight in Brazil—economist Sultan Mehmood of the New Economic School in Moscow and collaborators built JudgeGPT, combining OpenAI’s GPT-4 with a knowledge base of nearly 130,000 Pakistani judicial opinions and statutes. Offered in 2024 to 1,559 trial judges, roughly half the country’s justices, the tool used retrieval-augmented generation to query 128,292 opinions and 943 statutes, with footnotes linking to sources. MIT economics professor David Autor called the result credible, noting the productivity boost is likely to improve with wider use. Study coauthor Elliott Ash of ETH Zurich said attaching models to a search-and-verify tool helps fix hallucinations, though the researchers do not report hallucination rates. The team also trained 1,197 judges in six 90-minute Zoom sessions covering LLM limitations and bias risks. Mehmood said judges were enthusiastic, as commercial chatbots performed poorly on Pakistani legal queries. AI tools for judges are already rolling out in Brazil and India, but this study is the first major independent assessment of ongoing judicial AI use.