JudgeGPT — a Nationwide Randomized Trial of Generative AI for Pakistan's Trial-Court Judges
Pakistan
A randomized trial gave ~1,559 Pakistani trial-court judges (about half the bench) access to JudgeGPT, a GPT-4 tool searching 129,235 …
Croatia · Zagreb · See the Croatia profile · See the Zagreb profile
Evidence: Descriptive / self-reported Top 24% 73/100 · Ask Evidence Copilot about this practice
Since 1 January 2025, Croatia's ANON system has used AI/NLP to automatically anonymise and publish every court decision that concludes a case nationwide, on a searchable public portal, while legal commentators flag ambiguous redaction standards and new burdens on judges.
Croatia's Law on Amendments to the Courts Act (Official Gazette No. 36/24) mandated public disclosure of anonymised court decisions from 1 January 2025, with implementing rules set out in the Regulation on the Method of Anonymisation, Publication and Search of Anonymised Court Decisions (Official Gazette No. 134/24).
The ANON system aims to automatically anonymise, extract metadata from, thematically classify and publish every court decision that concludes a case, nationwide, on a searchable public portal.
ANON connects to the eSpis case-management system for new decisions and to the archived Supranova portal for decisions through 15 December 2024. First-instance decisions are retrieved, anonymised and published automatically between 15 and 60 days after dispatch; appellate and higher-court decisions go through additional judicial verification before publication. The public search portal, odluke.sudovi.hr, is operated by the Ministry of Justice, Public Administration and Digital Transformation, offering advanced search and export in multiple formats.
Since 1 January 2025, ANON has processed every concluding court decision nationwide under a binding legal mandate, though no independent figures on anonymisation accuracy or error rates have been published.
Croatian legal commentary (IUS-INFO) has raised open questions: the Regulation's redaction standards are described as 'framework-like' rather than precisely specified, criteria for judging which higher-court decisions are 'significant for judicial practice' remain unclear, additional manual anonymisation adds administrative burden for judges, and some courts have a documented history of publishing truncated decisions without marking them as shortened.
National / regional programmes
Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.
Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.
Where this practice's information was retrieved from, and when.
Pakistan
A randomized trial gave ~1,559 Pakistani trial-court judges (about half the bench) access to JudgeGPT, a GPT-4 tool searching 129,235 …
Kenya
Kenya's National Registration Bureau printed 1.7 million Maisha Namba ID cards in 2024 and cleared a 600,000-card backlog using biometric …
Russia
Moscow's public health department centrally benchmarks and deploys a marketplace of 50+ vetted computer-vision diagnostic tools as a mandatory "second …
United Kingdom
A Gemini-powered tool converts decades of paper planning records into structured digital data, saving councils an estimated 255 hours a …
Open full copilot Grounded in cited practices — always check the sources.