The new client’s outside counsel guidelines arrive attached to the engagement email: thirty pages on staffing, billing increments and travel, and on page twenty-two, a paragraph about generative AI. Your associates have been using a drafting tool for months. Nobody at the firm has written down which one, on what kind of account, or what goes into it. You read the paragraph twice and realize you do not know what the firm’s answer is.
The American Bar Association’s guidance on that question is Formal Opinion 512, issued on July 29, 2024 by its Standing Committee on Ethics and Professional Responsibility. It is the committee’s reading of the ABA’s Model Rules of Professional Conduct, not a New York rule, but New York’s own guidance, discussed below, runs in the same direction.
The opinion bans nothing. It takes seven duties every lawyer already owes and works out what each one means when a generative AI tool (software that writes text, summaries or drafts in response to a prompt) is in the workflow: competence, confidentiality, communication with the client, supervision, advancing only meritorious claims and contentions, candor toward the court, and charging reasonable fees. Its discussion groups meritorious claims and candor under one heading, which is why you will sometimes see it described as six. Two of the seven take real work: consent before client information goes into certain tools, and what happens to the hours the tools save.
Competence: understand the tool, not the engineering
The opinion says in terms that lawyers need not become experts in generative AI. What they need is a reasonable understanding of the capabilities and limitations of the tools they use. The sharp edge comes next: relying on the output uncritically is where competence fails. A draft from a tool is a draft: someone who knows the law reads it and checks every case, quotation and statute it cites.
Confidentiality: when you need the client’s consent
The confidentiality rule, Rule 1.6, governs what you put into these tools. Before you put information relating to a representation into a tool, you assess the risk that it will be disclosed to, or accessed by, people who should not have it. That is a judgment about a specific product and its terms, not a general stance.
Then comes the line that changes practice. Some tools are self-learning: they keep using what users type in to shape what they produce later, which means your client’s information could surface, directly or indirectly, somewhere else. For those tools, the opinion says the client’s informed consent is required before you input information relating to the representation. Informed consent is agreement the client gives after you have explained enough for them to understand the risk. And the opinion closes the obvious shortcut: adding general, boilerplate language to your engagement letters that purports to authorize AI use is not enough.
Keep the scope exact. For AI tools generally, the duty is the risk assessment; consent attaches to self-learning tools that will receive information relating to the representation. So you have to know which of your tools are self-learning, which means reading the terms of service rather than the marketing page (a vendor’s consumer and business versions can carry different terms). Where the answer is yes, you need a real conversation with the client and a note of it in the file.
Communication: when you have to say something
Rule 1.4, the rule on keeping a client informed, does not require you to announce every use of a tool. The opinion identifies when you must disclose or consult:
- when the client asks how you did the work, or whether AI was used;
- when the engagement agreement or the client’s outside counsel guidelines require disclosure;
- when your use of the tool is relevant to the basis or reasonableness of your fee;
- when the tool’s output will influence a significant decision in the representation.
The second item is the paragraph on page twenty-two, and your answer to it has to be true. And one more occasion comes from the confidentiality section above: before information relating to the representation goes into a self-learning tool, the client has to be told and has to give informed consent.
The engagement letter is a sensible place to describe how the firm uses these tools. But describing your practice there is disclosure, not the informed consent a self-learning tool requires.
Candor: checking was always the job
Two of the seven duties point at the court: advance only claims and contentions that have a basis, and be candid with the tribunal. A tool that invents a case does not change who signed the brief.
New York has now addressed this directly, though perhaps not in the way you have heard it described. By Administrative Order AO/75/2026, dated March 25, 2026, the Chief Administrative Judge added Part 161 to the court system’s rules, effective June 1, 2026. It applies in all courts of the Unified Court System, in civil and criminal cases. Its section 161.3 sets the statewide policy, which may not be what you expect: using AI to prepare a submission should not be prohibited, and lawyers and parties should not be required to disclose that they used it.
The review requirement people associate with Part 161 is not in section 161.3. It is in a model rule printed as Appendix A, and section 161.4 leaves each court free to adopt that model in its discretion. Where a court has adopted it, the filer must carefully review the submission and independently make sure it contains no fabricated or fictitious cases, statutes or other material. Where a court has not, Appendix A notes that the existing duties reach much the same place: signing a paper already certifies that it contains no false statement and no frivolous argument. So the answer to “does Part 161 make me certify my review?” is: only before a court that has adopted the model rule, and you find out by reading that court’s part rules.
The New York City Bar’s ethics committee covered the same ground for New York lawyers in Formal Opinion 2024-5, dated August 7, 2024. It says in its own notes that it is, in general, consistent with the ABA opinion.
Supervision: this is where the policy comes from
The opinion puts the obligation on management in plain words: managerial lawyers must establish clear policies on the firm’s permissible use of these tools, and supervisory lawyers must make reasonable efforts to ensure that the firm’s lawyers and nonlawyers comply with their professional obligations when they use them. A written AI policy is not a nice extra in that sentence. It is the thing the sentence describes.
Fees: the awkward one
A lawyer who bills by the hour bills the time actually spent; the opinion draws that from an earlier ABA opinion, Formal Opinion 93-379. If a tool turns four hours of document review into one, the client is billed for one.
A lawyer may not charge a client for time made necessary by the lawyer’s own inexperience, so you may not bill a client for learning a tool you will regularly use for clients: keeping up with your tools is part of staying competent. The opinion carves out the reverse case. Where a client asks you to use a particular tool that you do not otherwise use, billing the time to learn it may be appropriate.
What the opinion does not require
The opinion does not ban the tools, require client consent for every use, or require a particular vendor or product. What it does require is that someone at the firm has read the terms, decided what may go where, written it down, told everyone, and checked the output.
A worked example
An illustration, a composite rather than a client: a commercial firm of six lawyers.
- List the tools. There are three: two the firm approved, and one an associate has been using on a personal account. That one stops today.
- Read the terms of each. Suppose one learns from what users type on its consumer version but not its business version: move to the business version, or keep matter information out.
- Decide which tools are self-learning. Where one will see information relating to a representation, have the consent conversation with the client and note it in the file.
- Add a paragraph to the engagement letter on how the firm uses these tools.
- Write a short policy: approved tools, permitted uses, the checking rule, and who owns it.
- Fix the billing. If the tool saved three hours, the invoice shows the hour actually worked.
- Train everyone once, and record the date.
The firm’s AI counsel page describes this work for law firms: an audit that maps a firm’s tools and practices against Formal Opinion 512 and New York guidance, a written AI-use policy, and training for lawyers and staff.
Where to start
This week, send every lawyer and staff member one question: which AI tools have you used on client work in the last three months, and on what kind of account? Then pull the outside counsel guidelines for your five largest clients and search each for “artificial intelligence” and “AI”. Those two lists will tell you whether you have a consent question, a disclosure question, or neither.
This is general information about ABA Formal Opinion 512 and New York’s court rule on AI, not legal advice about your firm’s obligations. The guidance is still moving; check the current versions before relying on it.