What AI already does in legal work
Legal AI takes the part of the job where a lawyer spends hours reading and searching, not deciding. Thomson Reuters, in its Future of Professionals report of June 2025 (2,275 respondents), gives a targettarget variable. A target variable is the value a machine learning model learns to predict: whether a customer will leave, how much product will sell, how much a flat costs.: AI should save a professional 5 hours a week.
Those hours come from five tasks. Contract review: a check against the firm’s template and a list of risky clauses with their numbers. Search in case law and the firm’s archive, with a link to the document. Drafts of claims, lawsuits, motions and client letters from the firm’s templates. Case file analysis: a timeline, the evidence and a summary of hundreds of pages. Intake: the substance of the question, documents, deadlines and a booking for a consultation.
In all five tasks, AI prepares a draft or finds information. The lawyer keeps the legal assessment, the signature and the talk with the client.
Example: a lawyer receives a 40-page supply contract from the other side. The AI compares it with the firm’s template, lists the changed clauses on liability and jurisdiction, and proposes wording from the template. The lawyer reads a list of deviations, not the whole contract.
The market adopted the technology faster than firms wrote their own rules. The same Clio report says 87% of lawyers in large firms use AI and 71% of solo practitioners do, while 53% say their firm has no AI policyCompany AI policy. A company AI policy is an internal document that tells employees which neural networks they may use, which data they must not send to them and who is responsible for decisions made with AI. or they do not know of one.
Why public chatbots put client confidentiality at risk
A public chatbotchatbot. A chatbot is a program that exchanges messages with a person in a chat window: on a website, in a messenger or in an app. stores your requests on the provider’s servers under the provider’s rules, not the firm’s rules. For a lawyer this is a question of client confidentiality, not convenience.
On 17 February 2026, Judge Jed Rakoff (Southern District of New York) ruled in United States v. Heppner that a defendant’s exchange with a public AI service is not covered by attorney-client privilege. The court pointed to the service’s policy: requests and answers are collected and may be disclosed to third parties.
In Russia, attorney secrecy under Article 8 of the Federal Law on Advocacy covers any information linked to legal help for a client. Only the client can release a lawyer from it, not the service provider and not the lawyer.
A second layer comes from 152-FZ, Russia’s personal data law. Contracts, claims and letters almost always contain personal data: data of Russian citizens must be collected and stored in databases inside Russia (Part 5, Article 18), and since 1 March 2023 an operator must notify Roskomnadzor, the Russian data regulator, before any transfer abroad (Part 3, Article 12).
A lawyer may still use a public service. Send only what you would publish on the firm’s website: a general question about a rule, or text without names, amounts and details. The line runs along client information, not along the type of task. The article on attorney secrecy and AI covers the rule and the practice in detail.
How a private legal AI works
A private legal AIAI in legal work. AI in legal work means programs that help lawyers search laws and case law, review and draft contracts and sort large sets of documents. consists of a model, the firm’s document base and working rules, all inside a perimeter the firm controls. We install the model on the client’s server or in a cloud in Russia, so documents and messages do not leave the country.
It has four parts. An open language modellarge language model. A large language model (LLM) is a neural network trained on a huge amount of text. It writes, summarizes and analyzes text on request. that can be replaced without rewriting the rules. The firm’s knowledge base: templates, case positions, opinions and procedures, with every answer linked to a document. Citation checks: the AI marks any reference to a rule or act that is not in the base as unverified. Access rights and a log: the AI has its own account with rights only to its process, and every action goes to the log.
Citation checks matter because models invent case law. Damien Charlotin’s database lists 2,190 court cases with AI hallucinations as of 15 October 2026: 1,492 in the US, 232 in Canada, 76 in the UK and one in Russia.
Which tasks to start with
Pick the first task by two signs: it repeats every week, and its result is easy to check. Good examples are checking incoming contracts against a checklist, answers from the knowledge base, tracking procedural deadlines, first client requests and a case summary before a hearing.
Leave the legal position on a new dispute for the second stage: a mistake costs more there, and you cannot check the result quickly. Write the pilot metric into the contract, for example a lawyer’s hours per task before and after. The Thomson Reuters target helps set expectations, but count on your own cases; the time calculator helps with that.
A firm and an in-house team need different tasks: see the articles on AI for a law firm and AI for an in-house legal team.
How to protect client data
Personal data is processed on servers in Russia under 152-FZ, the Russian personal data law. Document text does not go abroad or to services without a data processing agreement. After launch, the system, the data and the access rights stay with the firm, and an administrator can revoke any AI access in a minute.
Your firm is the personal data operator, the party responsible for the data. So the firm needs a processing policy, client consent where the law requires it, and a contract with the vendor that includes a data processing section.
What a pilot costs and how it runs
Start with a pilot: one AI agentAI agent. By definition, an AI agent is a neural network that does not just answer but carries out a task step by step on its own: it looks for data, uses programs and checks the result. on one process of the firm, with the price, the term and the success metric fixed in the contract before the start. The next step is an AI departmentAI agent team. An AI agent team is a group of AI agents that run a company process together: each agent plays its own role, and a person sets the rules and checks the important decisions. of several agents on several processes, connected to the CRM, 1C (a Russian accounting system) and telephony. Prices and terms are on the AI department page.
For the pilot, the firm provides access to templates and the archive for the chosen process, examples of typical tasks for one month and one person for half an hour a day to settle disputed cases. If the model must run on your server, hardware is priced separately; prices are in the local AI cost calculator.

