What the Supreme Court approved
Order No. 59-P was signed by I.V. Krasnov, Chairman of the Supreme Court. The document sets the goal, principles, tasks and directions for introducing AI, and it was developed with regard to 243-FZ, Russia’s AI law, and the National Strategy for the Development of AI until 2030.
The tasks are listed in item 6: make access to justice simpler, make the court system more transparent, reduce the workload by automating routine work and cut the number of human errors.
Eight principles of introduction (item 5 of the Concept):
1. Protection of human rights and freedoms, with no discrimination in the design of solutions.
2. Auxiliary role: AI helps the judge, and the final decision always stays with a person.
3. Legality: AI is used only within procedural law.
4. Transparency, explainability and control: every interaction with an AI system is recorded.
5. Quality and reliability: source data comes from official publication sources.
6. Public order and ethics.
7. Confidentiality and personal data protection.
8. Technological independence: mainly Russian technology.
The introduction will be managed by the Center for Judicial Competence in AI, an expert group at the Supreme Court. Judges and court staff will be trained at the V.M. Lebedev Russian State University of Justice.
AI functions in courts
The Concept plans end-to-end use of AI: from filing documents to enforcing a court ruling and review by higher courts. The directions are listed in items 8.1–8.7.
8.1. Filing and first processing of documents: registration, checks of completeness and jurisdiction, grounds for leaving a claim without action or returning it, a state fee calculator, and a preliminary risk assessment for the applicant by standard category.
8.2. Classification of cases and their distribution among judges, with regard to workload and specialization.
8.3. Support during the hearing: a record by speech recognition with speaker identification, real-time translation, and a selection of rules and case law for the judge during the hearing.
8.4. Help with the case: analysis of positions and evidence, answers on the case materials with the source document named, case law search, a warning when a draft decision departs from case law, and reference recommendations on sentencing.
8.5. Draft court rulings: the structure and text of a draft decision, order or ruling based on the case materials, plus checks of spelling, style and logic.
8.6. Case law analysis: reports by case category, reasons for changes and reversals, repeated errors.
8.7. Collection and analysis of statistics, with analytical dashboards for court leadership.
Each solution goes through four stages (item 10): a process audit and a search for a solution, a pilot, an assessment of results by metrics, and scaling with user training. So the functions will appear one by one, not all at once.
What stays with the judge
The Concept’s goal is written with a condition: court efficiency grows while the judge’s key role in decisions is kept without exception. The principle of the auxiliary role repeats this: AI helps the judge and does not replace them.
The document rules out a forecast of the outcome of a specific case. A preliminary risk assessment for the applicant is given only by standard category and with an explanation of the reasons.
The risk matrix turns the principles into rules. Against weaker judicial independence, it sets a ban on “default choice”, a required justification when a judge rejects an AI recommendation, and interaction logs.
Against a breach of the right to a fair trial: the party’s right to know that AI is used, the right to challenge its conclusions and the judge’s required final role.
The Concept rates model hallucinations as a risk with high probability and very high impact. The measures are answers with a link to the source through a RAG architecture, a check of the result by the judge and a ban on public AI services.
The Center for Judicial Competence updates the risk matrix at least once every six months. The measures will be refined after the pilots, so the document is worth following.
The targets also leave control with the judge. By 2030, AI should prepare 30–50% of draft procedural documents, and judges should approve at least 80% of drafts without major edits: the draft is the machine’s work, the decision and the signature are the judge’s.
Experience of other countries
Other countries have already made similar decisions. In December 2022, the Supreme People’s Court of China required courts to have an AI support system by 2025 and rules for the whole process by 2030; decisions are always made by judges.
On 31 October 2025, the judiciary of England and Wales issued updated AI guidance for judges. Its rules: do not enter private information into public AI services, keep in mind the risk of invented answers, and take personal responsibility for materials issued in your name.
The EU AI Act classifies systems that help courts research facts and apply the law as high risk (Annex III, point 8(a)). In its clarification, the European Commission separates the functions: a draft decision is high risk, while case distribution and anonymization of decisions count as auxiliary administrative work.
The general line matches the Russian one: AI prepares the material, and the judge decides and answers for it. The dates and the degree of regulation differ; a detailed comparison is on the page about AI in courts worldwide.
What it means for lawyers
The dates are close. In the fourth quarter of 2026, the software and the procedure for placing AI services in the closed environment of the court system are set; in the first quarter of 2027, priority scenarios go through a pilot; in the second quarter of 2027, all courts get access.
1. AI will read documents first. Direction 8.1 provides for automatic checks of completeness, jurisdiction and grounds for leaving a claim without action, so formal errors will show up earlier.
2. References will be checked automatically. The risk matrix names filing AI-generated materials with false references as a separate risk and proposes automatic checks of references to rulings and regulations, disclosure of AI use and procedural sanctions.
3. A party has the right to know that AI is used and to challenge its conclusions. The document does not explain how this right will work in a case: procedural rules still have to set that.
4. Parties will get services too. The target for 2030: at least 30% of participants in a case use court services based on AI.
5. Standard cases will move faster. The goal for 2030: 20–30% shorter review times for simple cases and 15–20% fewer appeals and reversals.
The conclusion for a firm: the court will work with AI in a closed environment and check references. It pays for a lawyer to build the same in-house: AI on the firm’s documents, answers with a link to the source and a check before signing.
The requirement for Russian technology and a closed environment is a guide for a firm too: against leaks, the courts chose an isolated environment and a ban on sending data to public clouds. Clients’ personal data is processed on servers in Russia under 152-FZ, the Russian personal data law, and the model can be installed on the firm’s server.