The Assisted Judge

On 3 September 2026 the President of the Supreme Court of the Russian Federation, Igor Krasnov, signed Order No. 59-П. It approves the “Concept for the Introduction of Artificial Intelligence Technologies into Judicial Proceedings of the Russian Federation” and three annexes: a timetable, nine indicators to be reached by 2030 and a risk matrix ¹.

The indicators set the finishing line. By 2030 more than 95% of judges will have to use the courts’ artificial intelligence services on a permanent basis; at least 90% of hearing records will be prepared with AI; between 30 and 50% of draft procedural documents will be generated by the machine, and at least 80% of those drafts will have to be approved by judges without substantial changes ¹.

We read it in full, annexes included, to understand how the programme is built: which tools it provides for, at which stages of the proceedings, on what timetable, with which safeguards and with which measurable objectives. This article describes it as it is written. The comparison with European and Italian rules will be the subject of a forthcoming article.

In brief

• What it is. An organisational act of the President of the Supreme Court, setting principles, objectives, timetable and responsibilities for the introduction of AI into the courts ¹.
• The stages. Testing of the priority scenarios in the first quarter of 2027; access of all courts to AI services by the second quarter of 2027; quantitative targets for 2030 ¹.
• The areas. Seven lines of work, from the filing of documents to judicial statistics: automatic hearing records, case-file analysis, search for precedents, draft decisions, assistance in determining the sentence ¹.
• The guiding principle. AI is defined as an auxiliary tool: the final decision always remains with the judge, every interaction with the systems is logged and the results must be checked ¹.
• The pilot. Since May 2026 the courts have been testing Russian AI platforms to transcribe hearings; in August the project went from 10 to 21 regions ^{5, 6}.
• Training. Judges and staff will be trained by the Russian State University of Justice; the timetable provides for the training rules in the first quarter of 2027 ¹.
• Prosecution and the bar. The Prosecutor General’s Office has its own digital transformation programme to 2030 ¹²; for lawyers the Concept provides for a declaration of the use of AI in filed documents ¹, and the Supreme Court has already asked for a non-existent citation to be penalised ⁹.

The document

Order No. 59-П is dated 3 September 2026. It approves the Concept and entrusts its implementation to two offices of the Supreme Court, the office for the systematisation of legislation and the analysis of case law, and the office for informatisation, which will work with the Judicial Department at the Supreme Court, the body that administers the courts. The V. M. Lebedev Russian State University of Justice is to take it into account in training judges and staff. The order is to be published in the Bulletin of the Supreme Court and sent to the lower courts; the President of the Court has reserved to himself oversight of its implementation ¹.

As its legal basis the text cites the Constitution, the principles of international law and the laws on the judicial system, and states that it takes account of two acts: Federal Law No. 243-FZ of 26 July 2026 “on support for the development of artificial intelligence technologies” and the National Strategy for AI to 2030, approved by Presidential Decree No. 490 of 10 October 2019 ¹.

The declared aim is to increase the “potential” of the judicial system and the efficiency of the courts, “with the unconditional preservation of the key role of the judge” in decisions. There are eight principles: protection of rights and prohibition of discrimination; the auxiliary nature of AI, whereby the final decision always remains with a person; legality; transparency, explainability and controllability, with every interaction with the systems logged; data quality and verification of results; public order and ethics; confidentiality and protection of personal data; technological independence, meaning preferential use of Russian technologies ¹.

Formally, this is an organisational act of the President of the Supreme Court. It sets principles, objectives, timing and responsibilities for the judicial apparatus and indicates the method by which the tools will be introduced: in stages, with testing and evaluation before roll-out. The rules of procedure remain those of the procedural codes, which the document does not amend ¹.

The figures in the text

Annex 2 lists nine indicators to be reached by 2030 ¹.

Source: Annex 2 to Order No. 59-П ¹.

What AI will do, stage by stage

The Concept provides for a “cross-cutting” application across the whole life cycle of a case, from the filing of documents to the enforcement of the judgment and review by the higher courts. There are seven lines of work ¹.

Source: point 8 of the Concept ¹.

For each line of work the text states the aim: to reduce the routine work of court staff, to unify case law, to make decisions more predictable and to reduce the number of judgments reversed ¹.

The timetable

Annex 1 sets out the stages ¹.

Source: Annex 1 to Order No. 59-П ¹.

2030 is the horizon for the quantitative targets; adoption in the courts is planned for 2027. The text also sets out the method: each tool goes through four stages, analysis of the process to be automated, testing, evaluation with dedicated metrics and roll-out at scale ¹.

At the Forum of Chief Justices of the BRICS countries, held in New Delhi from 4 to 6 September 2026, Krasnov reported that the Russian judicial system already uses artificial intelligence to transcribe hearing recordings, analyse evidence and prepare document templates, and that in some regions four large language models are being tested, trained on the legal positions of the Supreme Court and on the decisions that guide case law in matters affecting citizens and businesses. The declared objectives are uniformity of case law, the elimination of logical contradictions and judicial errors, and more predictable justice; the Supreme Court, Krasnov added, has drawn up together with Russia’s major artificial intelligence technology groups the ethical principles for the use of neural networks in justice. In the same speech he proposed setting up within the Forum a permanent working group of the BRICS Supreme Courts on AI in justice and drawing up common ethical principles (see “The safeguards provided for in the text”) ³.

The Competence Centre

Implementation is entrusted to the Judicial Competence Centre for Artificial Intelligence, set up at the Supreme Court “in the form of an expert group”, with rules approved by the President of the Court. Its members are officials of the Court and of the Judicial Department, academics from the University of Justice, judges of the lower courts and, where necessary, representatives of the authorities, the scientific world and “technology partners” ¹.

The Centre has research, design and timetable-monitoring tasks; it is to engage with the authorities, the scientific community, technology partners and public organisations; its tasks also include raising the international standing of the Supreme Court. It must also update the risk matrix at least every six months ¹.

It is headed by Denis Kunev, a judge of the Supreme Court, who previously ran the IT and analytics directorate of the Prosecutor General’s Office ². Krasnov himself was Prosecutor General from 2020 to 2025; the Federation Council appointed him President of the Supreme Court by Resolution No. 399-SF of 24 September 2025 ⁴. Both bring to the project the experience of the digitalisation of the prosecution service (see “Prosecutors and lawyers”).

The safeguards provided for in the text

The document accompanies the functions with a set of safeguards, spread across the principles, the lines of work and the risk matrix ¹. We summarise them here, because they are the part of the text that defines the relationship between the judge and the machine.

• Human decision. AI is defined as an auxiliary tool: the final decision always rests with the judge, who checks the results produced by the systems.
• Traceability. Every interaction with the AI systems is logged, and the logs are used for subsequent analysis. The algorithms must be understandable and verifiable.
• Anchoring to sources. Against “hallucinations” the text prescribes a mandatory RAG architecture, in which every answer refers back to its source (law, judgment, document), and prohibits the use of public AI services; the models are to be stress-tested.
• Relationship with recommendations. The judge must not treat the AI’s suggestion as the “default choice”; if the judge decides not to follow a recommendation, reasons must be given. The text places this rule among the measures protecting the independence of the judge.
• Ban on profiling. Personalised analysis of judges on the basis of their decisions is prohibited; in analyses judges are anonymised.
• Rights of the parties. A party has the right to know that AI has been used in its proceedings and to challenge its conclusions.
• Data and infrastructure. The systems operate in a closed circuit, with no transfer of data to public clouds, with a preference for Russian solutions and with periodic security testing.
• Gradual approach. Roll-out in stages, training of judges and staff, involvement of judges in design; in the pilot phase most functions are voluntary.
• Rules for the parties. Anyone filing a document prepared with AI must declare it; references to laws and judgments are checked automatically; procedural sanctions are provided for false references.

Krasnov himself has explained the meaning of these safeguards. At the BRICS Forum he set out the values that a common framework document on AI in justice should express: the independence of the judge, the right of the parties to know the grounds of the decision, the prohibition on delegating the final judgment to the algorithm. He added that the infrastructure of justice must not depend on jurisdictions and technology platforms outside national control, and that the court is not a technological algorithm: it must remain centred on the person and use technology as a means, not as the end of justice ³. As early as June he had named the strengthening of the guarantees of fair justice as a principle of the digital transformation, recalling that responsibility for the final decision always remains with the judge ⁷.

Training judges

The Concept also provides for training. The order instructs the V. M. Lebedev Russian State University of Justice to take it into account in the refresher courses for judges and court staff ¹. The timetable places in the first quarter of 2027, before all courts gain access to the services, the rules for extending the tested scenarios and for training judges; the risk matrix lists training among the countermeasures against data leaks, the erosion of judicial independence and resistance within the judiciary ¹.

At international level, at the BRICS Forum Krasnov proposed reciprocal traineeships and training programmes on digital technologies for judges and court staff, to be launched immediately, without waiting for a formal agreement ³.

The operational form of these safeguards (regulations, instructions, any amendments to the codes) will be defined in the later stages: the Concept entrusts the Competence Centre with developing the use scenarios and the roll-out rules between the end of 2026 and 2027 ¹.

How the Court assesses the risks

Annex 3 contains the risk matrix: twelve entries, each with probability, impact and countermeasures ¹.

Source: Annex 3 to Order No. 59-П ¹.

The countermeasures are of three kinds. Technical: RAG architecture, closed circuit, Russian solutions, security testing, monitoring and retraining of models, A/B testing of new versions. Organisational: training, gradual roll-out, voluntary use in the pilot phase, involvement of judges, funding from several channels. Procedural: the party’s right to know and to challenge, declaration of the use of AI in documents, checking of references, sanctions. The two risks with the highest rating, that is high probability and very high impact, are model hallucinations and cyber attacks ¹.

The pilot in the courts

The Concept does not start from scratch. According to the rector of the Russian State University of Justice, Olga Tisen, AI has already been used for some years in several courts, and the best results in automating routine tasks have been achieved by the Moscow justices of the peace ⁵. In 2026 the pilot became a national project.

First phase: ten regions

On 26 May 2026 the Judicial Department at the Supreme Court, through its Information and Analytical Centre, launched a pilot project in the federal courts of ten regions: Karelia, St Petersburg and the Amur, Vologda, Ulyanovsk, Irkutsk, Leningrad, Tambov, Belgorod and Moscow regions. The AI platforms are placed within the internal circuit of the GAS “Pravosudie” system; judges and staff compare their functions for transcribing hearing recordings and identify other possible uses in office work. The declared aim is to reduce the workload of the court registries, shorten timescales and make the handling of cases more efficient ⁶.

On 10 June 2026, opening a round table in Belarus of Russian and Belarusian judges on the use of information technology in criminal proceedings, Krasnov explained that these Russian programs are meant to free court clerks from transcribing audio recordings, often hours long, and to allow judges to analyse the case-file materials quickly and to check that the texts of judgments are consistent with the legal positions of the Supreme Court. He also identified the new risks, namely cyber attacks, threats to confidentiality and the legal protection of digital evidence, specifying that the challenges are not only technical but also legal, organisational and ethical ⁷.

Second phase: twenty-one regions

On 26 August 2026 the Judicial Department announced the second phase: the project went from 10 to 21 constituent entities of the Federation. In the first phase more than 6,900 hours of recordings were transcribed, almost 288 days of continuous hearings; only Russian platforms listed in the national software register are used, always within the GAS “Pravosudie” circuit. Beyond the drafting of records, other uses are being studied ⁵.

In the background lies the workload of the courts. On 20 May 2026, at a seminar on the justices of the peace, Krasnov recalled that a justice of the peace decides on average 490 cases a month and that in 2025 the justices of the peace examined more than 27 million; he instructed the Judicial Department to draw up uniform minimum standards for the resources of the justices of the peace and unified approaches to the use of the courts’ information systems ¹³.

The analysis of AI cases

In parallel, on Krasnov’s instructions, a large-scale analysis is under way of cases concerning the use of AI technologies, the results of which will serve to ensure uniformity of case law and, if necessary, to amend legislation ⁷. The Court had announced it on 27 May 2026: it is the first analysis of its kind at national level, covers all types of proceedings and devotes a specific part to the protection of honour and reputation against defamatory content generated with AI, including deepfakes, and to the protection of intellectual property ⁸.

Prosecutors and lawyers: what rules?

The prosecution service

The prosecution service has its own framework, separate from the Court’s: the Concept for the Digital Transformation of the Prosecution Authorities to 2030, approved by Order No. 621 of the Prosecutor General of 11 September 2025, when the post was still held by Krasnov ^{4, 12}. It is the document in force, and it is of a different nature from the Court’s: it concerns the entire digital architecture of the prosecution service, and AI appears in it as one of the technical principles of the transformation, alongside optimisation, consolidation, reliability, standardisation, a single data model, unification, virtualisation, modularity, import substitution and information security ¹². The line of work on “high-tech supervision” provides for remote inspections, real-time statistics, integration of the prosecution’s systems with external state systems (operational records, records of wanted persons, registers of natural and legal persons), certified tools for big data and the use of AI in management, organisational and supervisory activity ¹².

The bar

For the bar, the most recent reference is the work of the Federal Chamber of Lawyers (FPA) on the rules adopted elsewhere: in July 2026 its adviser Olga Shvarts published a review of the guidelines on generative AI adopted in 2024–2025 by the bar bodies of several countries, all of them in the nature of recommendations ¹¹.

The Court’s Concept provides for the parties a declaration of the use of AI on filing, automatic checking of references and procedural sanctions for false references ¹. The Supreme Court has already applied this principle: by its ruling of 3 September 2026 No. 301-ЭС21-23756(4) it asked for a fine to be considered against the representative of a creditor who had cited a non-existent point of one of its plenary resolutions, stating that whoever prepares their position with the help of software does so at their own risk ^{9, 10}.

Summary by actor

whitecollarcrimes’ own elaboration of the sources indicated.

What changes for the parties in economic proceedings

Economic cases, with large case files, accounting documents, corporate chains and expert reports, are the area in which the tools described by the Concept will find their widest application: automatic summaries of the case file, chronologies, search for precedents by relevance, statistics on sentences in similar cases ¹.

For those who operate in Russia, or have cases in Russia, three changes are concrete or imminent: the declaration of the use of AI in filed documents, the automatic checking of citations and the possibility, provided for in the text, of knowing whether AI has been used in one’s own proceedings and of challenging its conclusions ¹. The prosecution’s Concept in turn provides for the integration of the prosecution’s systems with state databases and company registers ¹²: in proceedings for economic offences (fraud, Art. 159 of the Criminal Code; bribery, Arts. 290 and 291; money laundering) the prosecution’s preliminary analyses will be able to draw on a broader body of data.

What to watch in the coming months

The Concept is one of the most detailed national programmes published so far on the use of AI in the courts: it sets principles, stages, indicators and a risk assessment with countermeasures, and provides for updating it over time. Some deadlines make it possible to follow its implementation closely: by the end of 2026 the choice of technological solutions and priority use scenarios; in the first quarter of 2027 the results of the testing and the rules for training; in the second quarter of 2027 the access of all courts to the services; within six months of approval the first update of the risk matrix ¹. Also to be followed are the results of the second phase of the pilot in the 21 regions ⁵, the choices of the bar ¹¹, the translation into operational rules of the right to know of and to challenge the use of AI ¹ and, at international level, the proposal for a working group and common ethical principles for the BRICS countries ³.

In a forthcoming article we will compare this model with the rules adopted in Europe and in Italy.

Get each part by email as well, so it does not depend on any social network: whitecollarcrimes.it

Inga Santa Maria
whitecollarcrimes · 27 September 2026