MMOOZZGG.com

AI for lawyers

ABA Formal Opinion 512: the ethics duties of lawyers who use AI

5 min read7 sources

ABA Formal Opinion 512 put the ethics of lawyers who use AI into a regulator’s document: the American Bar Association published it on 29 July 2024. It names four duties for a lawyer who works with generative AI: competence, confidentiality, client communication and a reasonable fee. Below is what courts in England added and what the Federal Chamber of Lawyers of Russia (FPA) says.

On this page · 4 sections

Opinion 512: four duties

Opinion 512 does not ban AI. It applies the existing ABA Model Rules of Professional Conduct to it. The four duties:

1. Competence (Rule 1.1). The lawyer understands what the tool can and cannot do and checks its output: using AI without a reasonable independent check can breach the duty of competent representation.

2. Confidentiality (Rule 1.6). Before case information goes into a self-learning AI tool, the lawyer needs the client’s informed consent. A general standard clause in the engagement letter is not enough.

3. Communication (Rule 1.4). The lawyer tells the client about AI when the client asks or the contract requires it, before case information is entered, and when AI affects the fee or an important decision in the case.

4. Reasonable fees (Rule 1.5). On hourly billing, the lawyer charges the real time: for example, 15 minutes to enter data plus the time to review the draft. Time spent learning a tool the lawyer will use again and again is not billed to the client.

Flat fees are also in question. If AI let the lawyer finish a task much faster, charging the same amount as without it may be unreasonable under Rule 1.5. For a Russian firm Opinion 512 is not binding, but it gives a ready frame for its own policy and its contract with the client.

Related duties apply too. Under Rules 5.1 and 5.3, firm leaders set a clear AI policy and train staff, and Rules 3.1 and 3.3, on honesty toward the court, cover every citation in a court filing.

A separate duty concerns the terms of the service. The lawyer reads the tool’s terms of use and privacy policy, or asks a colleague or an outside expert to do it, to learn who will get access to the information entered.

England: responsibility for citations

In June 2025 the High Court of England and Wales heard two cases in a row. In Ayinde v Haringey, the filings contained 5 sources that do not exist; in Al-Haroun v Qatar National Bank, 18 invented cases.

Dame Victoria Sharp, President of the King’s Bench Division, reminded lawyers that they answer for the accuracy of their citations. The tool used to prepare a document does not move that responsibility: this is the same logic as the competence duty in Opinion 512. How such cases end in different countries is covered on the page about AI hallucinations in courts.

The English regulator is not against AI, though. On 6 May 2025 the Solicitors Regulation Authority (SRA) approved Garfield.Law, the first firm that provides regulated legal services only through AI: debt recovery claims up to GBP 10,000.

The conditions of approval are telling. The firm does not offer litigation, and named solicitors answer for the work. AI is acceptable when the process names a person who checks the result and answers for it.

Russia’s FPA: secrecy and consent

Under Article 8(1) of the Russian Law on Advocacy, professional secrecy covers any information related to the legal help an advocate gives a client. Under Article 8(2), an advocate cannot be questioned as a witness about facts learned while giving that help.

The wording of the law makes no exception for technical tools. Text an advocate pastes into a chatbot is still information related to legal help; the page “Attorney-client privilege and AI” covers this question in detail.

In December 2025 Sergey Zubkov, Vice President of the FPA, said the law does not forbid advocates to use AI, and an FPA working group is developing approaches. He linked the risk to sending sensitive data to servers abroad, which may lead to disclosure.

His advice matches Rule 1.6 of Opinion 512: warn the client and get the client’s consent. How to draft that consent is covered on the page about client consent to AI.

Ethics checklist

Six points that bring together Opinion 512, the practice of English courts and the FPA position:

1. Write down in the firm’s policy which AI tools may be used and for which tasks.

2. Read the terms of each service: where requests are stored, who can access them and whether they train the model.

3. Get the client’s informed consent before entering case information, in a separate document, not one general line in the contract.

4. Check every citation of a law or court decision against the original source before filing.

5. Bill the client the real time, including AI work, and do not bill time spent learning the tool.

6. Name a person responsible for supervision: who trains staff and who checks how assistants and contractors use AI.

Some points come from system design, not discipline. When the model runs on the firm’s server or in a cloud in Russia, the text does not go abroad or to services without a data processing agreement. The AI has a separate account with rights only for its own process, every action is logged, and the decision and the signature stay with the lawyer.

For the firm’s policy, Russia’s AI ethics code offers a local reference: its points on human oversight and responsibility are easy to carry into internal documents.

Sources and fact-checking

Checked by the editors on 16 October 2026.

  1. American Bar Association, 29.07.2024 (americanbar.org)
  2. American Bar Association, Formal Opinion 512, 29.07.2024 (americanbar.org)
  3. Legal Cheek, 09.06.2025 (legalcheek.com)
  4. Legal Cheek, 06.05.2025 (legalcheek.com)
  5. ConsultantPlus, Federal Law 63-FZ, Article 8, 31 May 2002 (consultant.ru)
  6. RAPSI, 25 December 2025 (rapsinews.ru)
  7. Data security in AI adoption

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