Should You Tell Clients You Use AI? What Rule 1.4 and ABA Opinion 512 Actually Require

Most lawyers asking whether they must tell clients about their use of AI are asking the wrong question. The duty is not to confess every tool you touch. It is to communicate what a client reasonably needs to know in order to make informed decisions, and to get consent before client confidences travel somewhere they should not.

No ethics authority in the United States requires a lawyer to make blanket disclosure of all AI use. Both the American Bar Association and the New York City Bar say the opposite: disclosure is sometimes required, sometimes advisable, and often unnecessary. The work is in telling the categories apart and handling them in the one document built for exactly this purpose, the engagement letter.

The rule that actually governs

The question of whether to disclose AI to clients is, at bottom, a Rule 1.4 communication question. Model Rule 1.4(a)(2) requires a lawyer to reasonably consult with the client about the means by which the client's objectives are to be accomplished. Rule 1.4(b) requires the lawyer to explain a matter to the extent reasonably necessary to permit the client to make informed decisions about the representation.

Notice what those provisions do and do not say. They do not require an inventory of your software. They require enough explanation for the client to make the decisions that are theirs to make. A spell-checker does not implicate that duty. A tool that shapes legal strategy, or that ingests the client's confidential file, can.

ABA Formal Opinion 512, issued July 29, 2024, is the governing national guidance. It declines to impose a general duty to disclose generative AI use in every matter. Instead it identifies specific circumstances in which disclosure or consent becomes mandatory, and leaves the rest to professional judgment. The opinion's own framing is that lawyers must make fact-specific inquiries rather than apply one rule to every engagement.

When disclosure or consent is required

Reading Opinion 512 together with Rules 1.4, 1.6, and 1.5, the situations that move from optional to obligatory are reasonably concrete.

• The client asks. If a client asks whether you used AI, or how, you answer candidly. This is the simplest trigger and the least ambiguous.
• The engagement agreement or outside counsel guidelines require it. Many sophisticated clients, particularly corporate and institutional ones, now include AI provisions in their outside counsel guidelines. If your agreement promises disclosure or restricts certain tools, that promise is binding regardless of what the ethics rules would otherwise demand.
• Client confidential information will go into the tool. This is the consequential one. Opinion 512 concludes that for self-learning generative AI tools, where inputs may be retained and used to train the model or may surface in outputs to others, a client's informed consent is required before inputting information relating to the representation. The concern is Rule 1.6 confidentiality, not communication for its own sake. The risk that one client's information could be disclosed, directly or indirectly, through a tool's later outputs is what creates the consent obligation.
• AI use bears on the fee. Opinion 512 ties this to Rule 1.5. If the use of an AI tool is relevant to the basis or reasonableness of your fee, that has to be communicated. More on fees below.
• The output will drive a significant decision. When a tool's output will materially influence a significant decision in the representation, the reasonable-consultation duty in Rule 1.4(a)(2) is in play, and the client should understand the role the tool played.

Outside those categories, Opinion 512 treats disclosure as a matter of judgment and, in some settings, best practice rather than command. A lawyer who uses a closed, enterprise tool that does not train on inputs, to draft an internal research memo the lawyer then independently verifies, is generally not under an affirmative duty to volunteer that fact.

What the NYC Bar adds

The New York City Bar Association reached a compatible conclusion in Formal Opinion 2024-5, issued August 7, 2024. It likewise declines to require blanket disclosure. A footnote in that opinion makes the point in plain terms: lawyers need not disclose generative AI that is routinely embedded in ordinary software, giving the example of Microsoft Word's autocomplete and grammar functions. No one expects a consent form for spell-check.

Where the NYC opinion is useful is on the texture of consent. It cautions that advance client consent to AI use that involves sharing client information is only effective if it is knowing, meaning the client actually understands the potential consequences of that information-sharing. A buried sentence the client never reads is not consent. The opinion also draws the line many lawyers blur, between open systems that share inputted information with third parties and closed systems that do not, and tells lawyers to know which kind they are using before they put a client file into it.

The practical upshot of reading the two opinions together: the trigger is rarely the abstract fact that AI was involved. The trigger is what the tool does with the client's information and what role its output plays in the representation.

The engagement letter is the natural home for this

If most of the obligation lives in confidentiality and informed consent, then the engagement letter is where it belongs. Handling AI client consent at intake, in writing, solves several problems at once. It satisfies the Rule 1.6 informed-consent requirement before any client data is entered into a tool. It sets expectations so a later question never feels like a discovery. And it converts an awkward mid-matter conversation into a routine term the client agreed to up front.

There is a drafting caution here that Opinion 512 makes explicit. Informed consent in this context is not the boilerplate most engagement letters already contain. A generic line authorizing the firm to use technology or vendors does not establish that the client understood the benefits, the specific risks, the categories of information that might be exposed, and how others might use that information. If you want consent to mean something, the language has to actually inform.

A workable engagement-letter approach covers a short list of points:

• A plain statement that the firm may use AI-assisted tools in providing legal services.
• A description, in client-readable terms, of the kinds of tools and what they do with information entered into them.
• Confirmation that the firm uses tools configured so that client information is not used to train external models, or, if that is not the case, specific consent to the alternative.
• A statement that a qualified lawyer reviews and remains responsible for all work product, so the client understands AI does not displace professional judgment.
• How any AI-related costs are billed, if they are passed through at all.

For clients with their own outside counsel guidelines, the letter should defer to or reconcile with those terms rather than contradict them.

Fees and billing, where lawyers get tripped up

The fee dimension deserves its own attention because it is where good intentions create exposure. Opinion 512 is direct that a lawyer billing hourly may bill only for time actually spent. If an AI tool lets you complete in one hour what used to take three, you bill the one hour. You cannot bill the pre-AI estimate, and the efficiency belongs to the client unless your fee arrangement genuinely provides otherwise.

On passing through the cost of AI tools as an expense, the opinion treats it like any other charge: it must be reasonable, and clients should not be surprised by it. General overhead is typically not separately billable. Where a specific tool's cost is charged to a matter, that should be disclosed and reasonable. Putting the billing approach in the engagement letter is the clean way to handle it, and it doubles as the Rule 1.5 communication the opinion contemplates.

A working checklist

Before a matter, and before any client information goes into a tool, run through this:

• Is the tool open or closed, and does it train on or retain inputs? If you do not know, find out before entering anything.
• Will client confidential information be entered? If yes and the tool is self-learning or open, obtain informed consent first.
• Does the engagement letter or any outside counsel guideline already require disclosure or restrict tools? Comply with the stricter term.
• Will the output drive a significant decision in the matter? If yes, consult the client about means as Rule 1.4(a)(2) requires.
• Does AI use affect the fee? If yes, address it under Rule 1.5, ideally in writing at intake.
• Has a lawyer independently verified every AI-assisted output before it leaves the office?

That last item is not a disclosure question, but it is the one that produces headlines. The widely reported sanctions in Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023), where the court imposed a $5,000 sanction on attorneys who filed a brief containing fabricated case citations generated by a chatbot, were not about failing to tell a client. They were about failing to verify. Disclosure obligations and verification obligations are different duties, and satisfying one does not satisfy the other.

The honest answer to the question

Should you tell clients you use AI? Sometimes you must, sometimes you should, and often you need not say anything beyond what a competent lawyer would say about any tool. You must obtain consent before client confidences enter a tool that could expose them. You must answer honestly when asked. You must address the fee implications. Beyond that, the prudent move for most small and mid-sized firms is not a one-off disclosure ritual but a standing engagement-letter term that handles consent, confidentiality, and billing at the start of every matter, in language a client can actually understand. That is less dramatic than a confession and considerably more defensible.

This is general information for lawyers and law-firm leaders, not legal advice, and it does not create an attorney-client relationship. The authorities are cited so you can read them yourself.