explainerAI in Courts AI Governance

May judges and court staff use AI in their own work?

The regulation map records 39 country answers on judicial AI use: binding rules, guidance, proposals and one case-law answer.

Edited and verified by Cognesio LLP2026-09-04

Researched with AI assistance · sources verified by Cognesio LLP · How this was made ↓

Yes, but the answer usually comes with limits on confidentiality, verification and adjudication. As of 4 September 2026, the judicial-use category records 10 binding country answers, 22 guidance answers, six proposals and one case-law answer. Among the non-positive records, 52 are provisional none records; one guidance record and one proposed record are also provisional.

How was the count made?

The count comes from the country YAML files in src/content/countries/. The script read each country’s judicial-use cell and grouped it by status. A country is counted once, even where the evidence comes from a regional court or a sub-national policy. That is why India appears as binding: the live policies are regional, but they bind judicial officers within those courts.

The category is narrower than court technology. It asks whether judges and court staff may use AI in their own work. A court’s procurement of transcription, translation or case-assignment tools belongs mainly in the court-own-use category unless the source also tells judges or staff what they may do with AI output.

Where is judicial AI use binding?

The binding group is Brazil, China, Colombia, India, Indonesia, Israel, Italy, Nepal, the Philippines and Spain.

The label does not mean each country has the same rule. India is binding at the regional court-policy level, with Kerala and Gujarat High Court policies and a Punjab and Haryana ban. Indonesia is binding only for the 2 September 2026 Supreme Court letter’s draft-template circulation rule. Italy and Spain use national legal instruments; Brazil and Colombia have judiciary-level governance instruments.

Where is it guidance or proposed?

Guidance answers are Argentina, Australia, Canada, Chile, the European Union, France, the United Kingdom, Hong Kong SAR, Croatia, Ireland, International bodies, South Korea, Malaysia, the Netherlands, New Zealand, Pakistan, Portugal, Rwanda (provisional), Sweden, Taiwan, the United States and Vietnam.

Proposals are recorded for Ecuador, Kenya, Morocco, Norway, South Africa and Ukraine (provisional). Mexico is the only country record marked case-law.

What do the documents actually say?

The tracker records each document in key-provision fields, and the primary texts support the map. The England and Wales judiciary guidance says: “Judicial office holders are personally responsible for material which is produced in their name.”

The New South Wales judicial guidelines say judges “should not use Gen AI in the formulation of reasons for judgment or the assessment or analysis of evidence preparatory to the delivery of reasons for judgment.” They also say: “Gen AI should not be used for editing or proofing draft judgments, and no part of a draft judgment should be submitted to a Gen AI program.”

The United States Courts annual-report page says the interim guidance “Cautions against delegating core judicial functions to AI, including decision making and case adjudication” and recommends extreme caution on novel legal questions.

The Mahkamah Agung letter annex states: “Dilarang menyebarluaskan rancangan kepada pihak di luar lingkungan tim, termasuk mengunggah ke media sosial, grup terbuka, laman berbagi berkas publik, atau layanan pengolah dokumen dan kecerdasan buatan;“

What is the common line?

The shared line is human responsibility. England and Wales makes the judge personally responsible for material in the judge’s name. New South Wales goes further for judgments: generative AI is out of bounds for reasons and for draft-judgment editing. The United States interim guidance allows experimentation, but not delegation of core judicial functions. Indonesia’s new letter is narrower still; it prevents draft judicial templates from leaving the working team through AI or document-processing services.

The result is permission bounded by task. Summarising, translation and administration are often allowed if the tool is approved and output is checked. Reasons, findings, credibility assessments, bail, sentencing and case adjudication sit at the hard edge of the rules.

Where has nothing been found?

The script counted 77 country records marked none and 14 marked unclear for judicial use. Because 52 of the none records are provisional, the absence finding should be read as a research status. It is not a final statement that those judiciaries have no internal policy.

What does it not cover?

Judicial-use rules address judges, tribunal members, clerks, researchers and court staff. They do not answer whether lawyers may use AI in filings, whether parties must disclose AI use, or whether courts may procure AI systems for case management. Those are separate tracker categories: verification duty, disclosure in filings and courts’ own AI deployment.