Why should a lawyer learn to work with AI instead of just using it?
Without training, a lawyer misses the errors an AI model makes. Yet 26% of legal professionals in the Thomson Reuters Institute survey already use generative AI, up from 14% a year earlier [2]. Responses were collected in early 2025 in the US, the UK, and Canada. Over 1,700 people from legal, tax, risk, and government work took part. Of all respondents across these four fields, 52% said their organization had no rules for using generative AI [2].
I spent twenty years leading software development for Walmart, Disney, Apple, and Microsoft, and I work with AI every day now. I write as a technology leader, not as a lawyer, so I name each legal rule below by its law, article, or source. In my experience, an untrained user takes the model’s confident tone for accuracy. In law, that mistake can cost a case and a reputation. So learn to check, not to write faster.
Which duties will move to the machine and which will stay with the lawyer is covered in our piece on whether AI will replace lawyers.
How do you check an AI answer so you never cite a fake case?
A lawyer opens every rule, case, and quote from the answer in a legal database before the text goes to a client or a court. The Law Society of England and Wales, in its guide to generative AI (updated 30 June 2026), calls this a risk to output integrity [3]. A model can produce an inaccurate or false result, and a person can then misread or misapply it.
The lawyer answers for that result, not the service: the client and the court see the lawyer’s signature, not the model’s name. I treat AI as a third opinion. It helps find and sort material, but the decision and the responsibility stay with a person.
- Find the rule in its current version and check the article and part number.
- Open the case by its number and make sure it exists and was decided as the model says.
- Compare the quote word for word with the document itself, not with a summary of it.
- Note in the draft what was checked, by whom, and when.
What happens without such a check is shown in our analysis of Mata v. Avianca, where lawyers cited cases ChatGPT had invented.
I use AI assistants to check facts and look for scientific papers, but I always judge their reliability myself.
In the book I wrote this about checking facts in general, not about law, but it fits a lawyer’s work as it is. Hence the main test of any program: with no exercise on finding a model’s error, it teaches writing, not working.
How do you protect client confidentiality while you learn?

For training, use only anonymizeddata anonymization. Data anonymization is the processing of information about people so that nobody can tell whom it concerns without extra information. or invented files. Client confidentiality and personal data stay protected even when a text is needed only for practice. When a lawyer uploads a file to a public service, the text goes to someone else’s servers. From then on, the lawyer no longer decides where it is stored or who reads it.
In Russia, a client’s personal data falls under Federal Law No. 152-FZ “On Personal Data” of 27 July 2006 [7]. Processing needs a legal ground (Article 6), and the data must stay confidential (Article 7). Uploading a file to a foreign AI service can be a cross-border transfer, governed by Article 12.
An advocate also keeps professional secrecy under Article 8 of Russia’s law on advocacy, Federal Law No. 63-FZ of 31 May 2002 [8]. In-house lawyers who are not advocates are bound by their employer’s confidentiality rules. These include Federal Law No. 98-FZ of 29 July 2004 “On Trade Secrets” [9]. A training purpose does not lift any of these rules.
In practice, an anonymized contract looks like this: the company name becomes Party 1, and the director’s surname becomes Representative. The contract number and amount become placeholder values. If a rare detail remains, such as a property address or a hearing date, change it too.
Detailed rules are in our piece on attorney-client privilege and AI and in the answer to whether a lawyer can upload client documents to ChatGPT. The mark of a good program here is simple: the teacher provides training documents and anonymizing rules, rather than asking you to bring your own.
How do you tell a good program from a sale of prompts?
A good program, by my rule, is more than half practice on legal documents. A live teacher checks the work and goes through the errors. Before paying, I would check six signs.
- Practice fills more than half the course
- The program was updated in the last six months
- There is an exercise on finding a model’s error
- There is a session on confidentiality and anonymizing
- A practicing lawyer is among the teachers
- Refund terms are written down before you pay
These thresholds are mine, and each has a reason. Six months, because services and what they can do change noticeably in that time. More than half practice, because the checking skill only grows on real legal texts. Refund terms before payment show whether the school stands behind the result.
I look at a program the way I look at any offer that asks for money. First I find out who answers for the result and how. A promise to master AI in one evening is worth nothing if nobody checks your work.
A ranking of courses with no stated method proves nothing. It does not show who set the places, by what criteria, and whether they are paid per click. General signs of an empty course are in our guide on how to choose an AI course. The checklist above adds the legal part.
Does a lawyer need an AI training certificate?
No certificate is mandatory in the EU. The European Commission’s explainer (updated 20 October 2026) says Article 4 of Regulation (EU) 2024/1689AI Act. The EU AI Act is Regulation (EU) 2024/1689 on artificial intelligence. It sorts AI systems by level of risk and sets requirements for each level., the AI Act, needs none [1]. The duty to train staff in AI has applied there since 2 February 2025. An internal record of training is enough [1].
Article 4 requires providers and deployers of AI systems to build the AI literacyAI literacy. AI literacy is the ability to use artificial intelligence with benefit and without harm: to understand what it can do, where it makes mistakes, which information should not be trusted to it and how to check an answer. of their staff. They must take into account people’s knowledge, experience, and where the system is used. Supervision and enforcement by national authorities have applied since 2 August 2026 [1].
Regulation (EU) 2026/1744 of 8 July 2026, the Digital Omnibus on AI, replaced Article 4 with a new text [10]. Providers and deployers must take measures to help develop the AI literacy of staff and others who operate or use AI systems for them [10].
Regulation (EU) 2026/1744 says this duty does not require providers or deployers to guarantee any specific level of AI literacy for any individual [10]. Under Article 4(2), the Commission and Member States support providers and deployers, especially small and medium-sized firms, and the Commission publishes practical examples [10].
The rule also reaches firms outside the EU, the Commission’s explainer says [1]. It applies when their AI system is placed on the EU market, used in the EU, or its use affects people there. Under the same explainer, a penalty is more likely when an incident is proven to come from a lack of training. For a firm with EU clients, this is a reason to keep a record: who learned what, and when.
Such a record fits in one table: employee, date, topic, hours, and the exercise used to check the result. It also covers a client who asks how the firm prepares its people to work with AI.
In Russia, general rules on AI ethicsAI ethics. AI ethics is a set of principles for building and using artificial intelligence so that it does not harm people: fairness, transparency, privacy and human responsibility. are collected in the Code of Ethics in the Field of AI, a national code. It is advisory, and joining it is voluntary. According to the Code’s website on 20 November 2026, more than 1,400 organizations have signed it, including over 90 foreign companies [5]. It creates no duties with penalties, so a certificate is worth something only if checked work stands behind it.
How long does training take, and where do you start?

Start with one recurring task from your own practice and two weeks of hands-on learning around it, not with a long program. In my experience, the skill holds up only through regular use. In the 2025 Thomson Reuters Institute survey, 72% of current generative AIgenerative AI. Generative AI is a family of neural networks that create new content: text, images, music, video and program code, from a description in plain language. users in the report work with it at least weekly [2]. Over 40% of them use it once or several times a day [2].
- Pick a task that comes up every week: a summary, a contract review, or a reply to a claim.
- Time yourself doing it by hand on three to five files with client details removed.
- For two weeks, do it with the model and check every answer with the steps from the section on checking.
- Compare the time and the number of errors caught with the manual work, and write down the result.
An example with made-up numbers: summarizing a 20-page court decision by hand costs 50 minutes. If the model and the check bring that down to 30, the 20 minutes saved on one summary is your case for your manager. At 15 such summaries a month, that is 5 hours.
I do not name a length for the whole course of study. It depends on the job you choose and on your weekly schedule. After two weeks you will have not a certificate but your own number: how much time the model saves you on one summary or review.
Should you learn on your own or train the whole firm?
One lawyer needs the skills; a firm also needs usage rules, a training record, and a person responsible for both. In the Thomson Reuters Institute report of 15 April 2025, generative AI training was open to 40% of law firm respondents [2]. For corporate legal staff, it was 41% [2]. Across all respondents, 31% said training was available in 2025, up from 19% in 2024 [2]. Available differs from received: 64% had received no specific training [2].
| Factor | One lawyer learns | The whole firm learns |
|---|---|---|
| What is needed | The skills from the table above | Skills, usage rules, and a training record |
| Who is responsible | The lawyer | A named person in charge |
| Training level | One level, for your own tasks | Different levels for different staff |
| What to show a client | Your own answer check | The firm’s rules on working with AI |
The European Commission (explainer updated 20 October 2026) openly allows different training levels for different staff [1]. The reason is that the AI Act asks organizations to consider people’s knowledge, experience, and background. For a firm, this means an assistant needs the rules and the answer check. A partner whose signature goes on filings needs the full set of skills.
Clients still know little about how their lawyers work. In the same 2025 study, 71% of law firm clients did not know whether their firm used such tools [2]. Rules written in advance answer that question before anyone asks it. A lawyer’s duties when working with AI are covered in our piece on the ethics of AI use by lawyers.
What should you learn besides the tools?
Besides the tools, a lawyer needs to know why the model makes errors and how the law regulates it, whatever the service is called. D. V. Shibaev described an Artificial Intelligence Systems course for law students [4]. His article appeared in the Courier of Kutafin Moscow State Law University (MSAL), 2024, issue 10. The course treats AI as a subject of study, a learning tool, and an object of legal regulation.
Shibaev proposes refining Russia’s educational standards, and in his words the course’s practical tasks give future lawyers digital knowledge and skills. The European Commission’s explainer (updated 20 October 2026) gives advice even to people with technical training [1]. They should check whether they know the legal and ethical sides of AI.
Understanding why the model errs means knowing one simple thing: it builds an answer from likely continuations of text. It does not find the answer in a database of laws. So a case number in an answer can look plausible and still not exist. A lawyer who understands this checks citations out of habit, not after the first unpleasant surprise.
The terms this conversation needs, from neural network to anonymization, are collected in our 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. glossary. Practice on a service your firm’s rules allow. Which options a firm has and how to compare them is covered in our legal AI tools comparison.
My book Kak ne stat bankrotom!? (How Not to Go Bankrupt) argues that AI evens the chances of a small business and a large corporation. I end it this way: I may not become a billionaire. But my chance of becoming a millionaire grows if I stay flexible and educated and learn to use artificial intelligence well. For a lawyer the conclusion is the same: the winner is the one who keeps learning, not the one who learned once.
Where can you learn for free, and what should you expect from a free course?
A free introductory course gives you the basics and the vocabulary, but it does not check your work on legal texts. A good first step is our free AI course: it needs no sign-up and gives a general foundation. From an introductory course, honestly expect three things: vocabulary, an idea of how a model builds an answer, and first simple drills.
I believe knowledge about AI should be open to everyone, like water and air, so a free course is a normal start for me. I do not name schools: choosing by criteria is safer than choosing from a list. There is one test: what you can do on your own after the course, and who will check that it worked.
After the introductory course, move to the two-week plan from the section on where to start. It will show whether you need a paid program.



