Artificial Intelligence in Judicial Proceedings

The use of AI tools in legal proceedings is becoming increasingly common. They are used to prepare court documents, analyse case materials, and search for relevant case law. At the same time, courts and regulators in different countries are increasingly introducing rules governing the use of AI, partly in response to cases where AI-generated references to non-existent court decisions and case law have resulted in additional costs or delays in court proceedings.

What has changed in Russia

Russia has adopted a framework law on the use of artificial intelligence, with its main provisions taking effect from 1 March 2027. The law introduces the concepts of sovereign and national AI models, whose developers may qualify for state support, and establishes a general principle: the technology itself is not recognized as a subject of liability. If a specialist submits a document to a court or signs off on a report prepared with the help of a neural network, liability rests with the person, not the program.

The law does not introduce direct new obligations for lawyers or accountants as such, but experts recommend that companies already put internal rules in place for using such tools, since liability will fall on the employee or the organization regardless.

Why Courts Have Started Paying Attention to AI

In May 2026, an arbitration court in one of Russia’s judicial districts held a legal representative liable and imposed a RUB 50,000 fine for citing non-existent Supreme Court rulings that had been generated by AI and not verified before the documents were filed.

Shortly afterwards, the Plenum of the Supreme Court established an obligation for parties to proceedings to disclose the use of AI when presenting factual information. At the same time, pilot projects using AI in the internal operations of courts were launched in ten regions.

How neighboring jurisdictions are addressing the same issue

Kazakhstan went further, adopting a universal law on artificial intelligence back in early 2026 with mandatory requirements for risk management, audit, and the labeling of synthetic content. In May, all courts in the country completed the rollout of a digital judicial assistant, a system that selects relevant case law and forecasts the likely outcome of a case, though the final decision still rests with the judge. Belarus does not yet have a dedicated law: existing rules on personal data and copyright are being applied instead, while an interagency working group is preparing a draft based on the model AI law adopted at the CIS level.

Despite the differences in approach, all three jurisdictions converge on several basic principles:

• artificial intelligence is not recognized as an author and does not become an independent subject of liability;
• the output of a neural network must be verified against primary sources rather than taken on faith;
• transferring confidential data to AI services without the client’s consent is prohibited;
• the final legal or judicial decision remains with a human being in every case.

Responsibility for Neural Network Errors

A separate question now actively debated in the legal community is who bears the cost when an AI error leads to a lost case or a drawn-out process. There is no single answer yet: if outside counsel prepared the document, liability typically falls on them as the service provider, but if the error entered the case file through an in-house employee, the issue becomes an internal one – professional liability insurance, disciplinary measures, and the sign-off process before a document is filed.

The legal community is also discussing broader, systemic risks: how far the use of AI in preparing procedural documents, and even in judges’ own review of case materials, could affect the quality of justice itself if verification of the output turns out to be merely a formality. For now this remains more a subject of debate than a settled rule, but the direction suggests the regulation will keep being refined over the next few years.

What not to do when working with neural networks

The law does not directly prohibit employees from uploading data to AI services for work tasks, but this is constrained by other rules – on personal data and on trade secrets – as well as by the terms of use of the particular service. In practice, lawyers and accountants are advised to follow a few simple rules:

• do not upload employees’ or clients’ personal data, bank details, or other confidential information to public neural networks;
• do not share unpublished contracts, internal reporting, or materials from tax audits;
• use corporate accounts rather than personal ones, so the company retains control over the history of queries;
• keep drafts and intermediate versions of documents, since these can help demonstrate a genuine creative contribution if an authorship dispute arises;
• log cases where a neural network made an error or produced unreliable information, so internal usage rules can be adjusted in good time.

What this means for business

Companies that work with outside counsel or maintain an in-house legal department should already be putting internal rules in place for the use of neural networks: keeping a register of the tools used, relying on corporate rather than personal accounts, and requiring every citation to case law or a statutory provision to be checked against the primary source before it is filed with a court.

A mistake of this kind can cost more than a fine – it can damage credibility in a specific case, since an unreliable reference to a non-existent court ruling can undermine confidence in a party’s position as a whole. When choosing outside counsel, it is worth asking how the firm controls the use of such tools, as this is increasingly becoming part of the due diligence companies apply when engaging legal support.

Timing is also worth keeping in mind: the main provisions do not take effect until 1 March 2027, and information systems that were already using AI by that date get a transition period until 1 September 2032 to bring their models into line with the sovereignty requirements. That gives businesses time to build internal processes methodically rather than in a rush, but it makes sense to start now, while the rules are still being shaped and industry associations still have a chance to influence them.