MMOOZZGG.com

AI for lawyers

Law firm AI policy for a firm or legal department: a template by section

6 min read5 sources

A law firm AI policy decides which AI tools lawyers use, which information they must never send to them and who checks the result. According to Clio Legal Trends 2025, 79% of lawyers use AI, and 53% say their firm has no AI policy or they do not know of one. Below is a template by section for a law firm and a company’s legal department.

On this page · 5 sections

Why a firm needs a policy

Lawyers already work with AI, with or without a policy. According to Clio, that is 87% of lawyers in large firms and 71% of solo practitioners, and a ban without an approved tool only moves the work to personal phones and subscriptions.

A policy settles three questions that each lawyer would otherwise answer alone: which tool may be opened, what may be pasted into it and whose signature stands under the result.

In Formal Opinion 512 of 29 July 2024, the American Bar Association tied work with generative AI to four duties of a lawyer: competence, confidentiality of client information, communication with the client and a reasonable fee. It is convenient to build a policy on the same structure.

In Russia, Sergei Zubkov, vice president of the Federal Chamber of Lawyers, said in December 2025 that the law does not forbid attorneys to use AI, but the client must be warned and must give consent. A firm’s policy turns this position into a procedure for every employee.

The template by section

A template: adapt it with the firm’s own lawyer.

1. Purpose and scope. The policy sets the procedure for the use of artificial intelligence systems by the firm’s partners, lawyers, assistants and trainees in providing legal services and in internal work.

2. Approved tools. Employees use only tools from the list approved by the managing partner. Personal accounts in AI tools are not used for work documents.

3. Classes of information. Information is divided into public, internal and protected. Protected information: attorney-client privilege, personal data of clients and third parties, and clients’ trade secrets.

4. Client consent. A client’s protected information is processed in an AI tool only with the client’s consent, obtained in the way the firm has set.

5. Checking the result. The lawyer checks each reference to a law or court ruling against the original source before the document goes to the client, the court or the counterparty.

6. Signature and responsibility. The lawyer who signed or sent a document prepared with AI is responsible for it.

7. Log. The firm keeps a log of requests to approved tools: who worked with AI, when and on which case.

8. Incidents. If protected information enters an unapproved tool, the employee reports it to the responsible person on the day it is found.

9. Training and review. New employees get a briefing before they receive access to the tools, and the responsible person reviews the policy at least once a year.

Allowed and prohibited data

Draw the line by tool, not only by document. In an approved tool in a closed network, case materials are acceptable too; in a public chatbot, only public information.

A sample table for an annex to the policy:

1. Public information: texts of laws, published court rulings, general questions on how a rule is read. Any approved tool fits.

2. Internal information: the firm’s templates, guidance materials, anonymized fragments of documents. An approved tool with a data processing agreement.

3. Protected information: case materials, correspondence with the client, personal data, deal terms. Only a model in a closed network in Russia, and only with the client’s consent.

Why a public foreign service does not fit the third group: before a cross-border transfer of personal data, the operator must notify Roskomnadzor, the Russian data protection regulator (part 3 of article 12 of 152-FZ, Russia’s personal data law), and data of Russian citizens is collected and stored in databases inside Russia (part 5 of article 18). The fine for a company under part 8 of article 13.11 of the Code of Administrative Offenses is up to $72,000.

The Federal Chamber of Lawyers gives the same warning: sending sensitive data to foreign servers may lead to its disclosure. For an attorney, this is a question not only of 152-FZ but also of privilege.

Checking the result

AI prepares a draft, and the decision and signature stay with the lawyer. In Formal Opinion 512, the ABA treats checking the result as part of the duty of competence: the lawyer is responsible for a filed document even if a tool wrote the text.

What to write in the policy about checks:

1. Each reference to a law, a court ruling and a case number is checked against the original source, not against the answer of another AI tool.

2. Quotes from case documents are checked against the page of the source document.

3. Amounts, deadlines and penalties are recalculated by hand or in a spreadsheet.

4. The file keeps a note of who checked the document and when.

Cases of AI hallucinations in courts show why this rule is a separate clause in the policy: lawyers who filed references without checking them get fined.

The ABA also ties the duty to inform the client and keep the fee reasonable to AI. If a tool cut the work on a document several times, a bill based on the old hours will need an explanation to the client.

Responsibility and the log

The responsible person is appointed by order: a partner or the head of the legal department. This person keeps the list of approved tools, takes incident reports and reviews the policy.

A log is needed for two reasons. It shows the client and an inspector that information was processed only in an approved tool, and it helps find the source if a document with an error went to court.

In a closed network, the system keeps the log itself: AI has a separate account with rights only to its own process, and its actions are logged. The firm sees which documents AI read and what it prepared.

The client consent that clause 4 of the template refers to is covered separately on the page about client consent to AI. An example of a strict rule, under which a person confirms each AI action, comes from the Bank of Russia in its AI security recommendations for the financial market.

Sources and fact-checking

Checked by the editors on 16 October 2026.

  1. 2Civility (review of the Clio report), 2025 (2civility.org)
  2. American Bar Association, 29.07.2024 (americanbar.org)
  3. RAPSI (in Russian), 25 December 2025 (rapsinews.ru)
  4. Brace Law Firm (brace-lf.com)
  5. Data security when introducing AI

Found a mistake? Tell us

Publications

Mike Mozg

Agentic AI use cases at AT&T: nearly 500 processes now run on agents

AT&T no longer uses AI only to help staff work faster: it is rebuilding whole business processes around agents. SiliconANGLE reported on 14 November 2026 that nearly 500 agentic AI use cases at AT&T are already in production. Here is how the company rebuilds them, where a human still checks the agent, and what other large companies can take from it.

News4 min
Contact us