An engagement letter AI clause tells the client whether and how your firm uses AI on their matter, and with what safeguards. Which version you need turns on one fact: does client information go into the tool? If not, a short disclosure will do. If it goes into a tool the firm controls, you need a specific disclosure. If it goes into a self-learning tool, ABA Formal Opinion 512 requires the client's informed consent first.
Below are three copy-ready clauses, then a client-side clause most templates leave out, a fee line and a script for existing clients. Fill every bracket. These are drafting aids, not legal advice; your state's rules control.
Which engagement letter AI clause do you need?
Opinion 512 does not treat all AI use alike, and neither should your letter. It reasons that "because many of today's self-learning GAI tools are designed so that their output could lead directly or indirectly to the disclosure of information relating to the representation of a client, a client's informed consent is required prior to inputting information relating to the representation into such a GAI tool." Uses where no such information goes in need no consent: the opinion's example is idea generation that "does not require inputting information relating to the representation."
For other tools, the opinion does not say consent is automatic. It says lawyers "must evaluate the risks that the information will be disclosed to or accessed by others outside the firm" and inside the firm, and that this analysis "will be fact-driven and depend on the client, the matter, the task, and the GAI tool used to perform it." (ABA Formal Op. 512, full text.) So the tiers below turn on what your firm does. Our Opinion 512 walkthrough covers the other duties.
| Tier | What the firm does | What the letter needs |
|---|---|---|
| A | No information relating to the representation enters any AI tool | Short disclosure of the firm's position |
| B | Client information goes into a firm-controlled, non-training tool | Specific disclosure of tool class, data handling, human review; consent where your state's rule or the facts call for it |
| C | Client information goes into a self-learning or public tool | Specific informed consent under Opinion 512, or avoid the use |
Tier A: no client information goes into AI
Use this if AI at your firm means brainstorming or tools that never see a client name, fact or document.
Use of artificial intelligence. The Firm does not enter information relating to your matter into generative artificial intelligence tools. The Firm may use such tools for general purposes that involve no client information, such as brainstorming or generic drafting aids. A lawyer reviews all work product and remains responsible for your matter. If this changes, the Firm will tell you before any information relating to your matter is entered into such a tool. You may ask us at any time how AI has been used on your matter.
If you cannot keep the promise to tell the client first, use Tier B.
Tier B: client information goes into a firm-isolated, non-training tool
This is the common case for firms that use AI on client work: a legal research or document tool that holds the firm's files. The disclosure has to be specific: name the tool class, say where the data lives and what the vendor cannot do with it, and say a lawyer reviews the output.
Use of artificial intelligence. The Firm uses [name of tool or tool category], a generative AI tool, to assist with [legal research, document review, drafting]. Information relating to your matter, including documents you provide, may be entered into this tool. The Firm has reviewed the tool's terms of use and privacy policy. Under those terms, [your firm's documents are stored in an environment isolated from other customers' data], [the vendor does not use your information to train its models], and [data is deleted on request / at the end of the engagement]. The Firm does not enter your information into AI tools that do not meet these conditions. A lawyer of the Firm reviews all AI-assisted work, verifies every legal authority against its source before it is relied upon or filed, and remains responsible for the advice you receive. The benefits are [faster, more thorough research and review]. The risks include [the possibility of error, which review is designed to catch, and the residual risk of unauthorized access inherent in any electronic system]. Please tell us if you object to this use or wish to impose limits on it. [If required by your state: By signing below, you consent to this use.]
Every bracketed data-handling statement must be true on the day you sign and checkable in the vendor's contract. Opinion 512's baseline is that "all lawyers should read and understand the Terms of Use, privacy policy, and related contractual terms and policies of any GAI tool they use." California's 2026 guidance adds that "reasonable efforts require more than reliance on generalized marketing assurances."
Florida's Ethics Opinion 24-1 (January 19, 2024) recommends that "a lawyer obtain the affected client's informed consent prior to utilizing a third-party generative AI program if the utilization would involve the disclosure of any confidential information." California's guidance says that, "as a general matter, a lawyer must not input any confidential information of the client into a generative AI solution that may present material risks to confidentiality or security, absent informed client consent." California's test turns on the tool's risk, which your Tier B facts are meant to address. Florida's turns on whether confidential information goes to a third party at all (it excuses consent only for uses that disclose nothing to one), so its recommendation still reaches a vendor-hosted Tier B tool. Check your state in our state-by-state guide (California has its own page). When in doubt, include the consent sentence.
One example of a checkable bracket: CaseRead keeps each firm's documents in that firm's own isolated database schema. No tool satisfies Rule 1.6 for you, though; the duty stays with the lawyer.
Tier C: self-learning or public tools
Here Opinion 512 is blunt: "To obtain informed consent when using a GAI tool, merely adding general, boiler-plate provisions to engagement letters purporting to authorize the lawyer to use GAI is not sufficient."
What informed consent does require, per the opinion: "the lawyer's best judgment about why the GAI tool is being used, the extent of and specific information about the risk, including particulars about the kinds of client information that will be disclosed, the ways in which others might use the information against the client's interests, and a clear explanation of the GAI tool's benefits to the representation."
A clause that meets that standard belongs on a separate consent page:
Consent to use of [tool name]. I ask your consent to enter the following information relating to your matter into [tool name]: [specific categories, for example, the facts of the dispute and the contract at issue; not Social Security numbers or medical records]. I want to do this because [specific reason and benefit]. [Tool name]'s terms permit [the vendor to retain inputs for __ days / to use inputs to improve its models / human review]. The risk is that information entered may later be disclosed to or accessible by other users of the tool or the vendor, in a way that could be used against your interests. Alternatives: I can do this work without the tool, [at greater cost / more slowly]. You may decline, and declining will not affect my representation. Consent: ____________ Date: ________
For most small firms the better answer is to move the work to a Tier B tool. Our guide on uploading client documents to ChatGPT explains why consumer tiers rarely qualify.
The client-side clause most templates skip
The clauses above govern what the firm does. This one governs the client, which is where one recent case went wrong. In United States v. Heppner, No. 25-cr-503 (JSR) (S.D.N.Y. Feb. 17, 2026), Judge Rakoff held that documents a criminal defendant generated with a consumer AI tool, on his own initiative, using information he had learned from his lawyers, were neither privileged nor work product. As the court's memorandum records, the defendant acted of his own volition, without direction from counsel. The lawyers did nothing wrong; the client created the exposure. More in does AI waive attorney-client privilege.
So tell the client, in writing:
Your own use of AI tools. Communications between you and the Firm are confidential and may be privileged, and you can lose that protection by sharing them with others. Please do not enter our communications, documents the Firm sends you, or the facts and strategy of your matter into any AI chatbot or similar tool, including consumer products, without first speaking with us. Doing so could make that information available to the tool's provider and could affect whether it stays confidential. If you have already done so, please tell us. If you would like to use AI to help you understand your matter, ask us and we will discuss options.
The last two sentences matter: a bare prohibition is easy to ignore, and an offer to talk gives the client somewhere to go. The clause promises no protection; it records that you warned the client.
Fees: say how AI cost is handled
Overhead stays in your rate. Under Opinion 512, if "a lawyer uses a GAI tool embedded in or added to the lawyer's word processing software to check grammar," that is overhead. But when a third-party service "charges the lawyer for using the tool on a per-use basis," it "would ordinarily be reasonable for the lawyer to bill the client as an expense for the actual out-of-pocket expense incurred." California's 2026 guidance says per-matter AI charges must be disclosed in the fee agreement, reflect actual cost, and carry no markup without the client's informed written consent. Florida's 24-1 says a lawyer "may only ethically charge a client for the actual costs incurred on the individual client's behalf and must not duplicate charges that are already accounted for in the lawyer's overhead."
AI and fees. The Firm bills for the time its lawyers and staff actually spend on your matter, including time spent directing AI tools and reviewing their output. The Firm does not bill for time saved by AI tools, or for time spent learning them. The Firm's general AI subscriptions are part of its overhead and are not charged to you. [If applicable: Charges imposed by a third-party AI service for use specifically on your matter will be billed at the Firm's actual cost, without markup.]
Flat fees and hourly mechanics are in billing for AI-assisted work.
A script for existing clients
For current clients, send a short note when a tool begins touching their file:
"We have started using [tool category] to help with research and document review on matters like yours. Our lawyers check everything it produces, and responsibility for your matter has not changed. Please reply to this email if you have questions or would like us not to use it on your matter. [If consent is required in your state: Please reply 'I consent' to confirm.]"
Keep the reply in the file. If a self-learning tool is involved, use the Tier C consent page instead. Match every clause to your written AI policy.
Frequently asked questions
Is a general AI sentence in my engagement letter enough?
Enough for disclosure when no client information goes into AI, not enough for consent when it does. ABA Formal Opinion 512 says that to obtain informed consent, merely adding general, boiler-plate provisions to engagement letters purporting to authorize the lawyer to use GAI is not sufficient. The consent has to explain the specific tool, the information involved, the risks, and the benefits.
Should the clause name the AI tool?
Naming the tool is good practice where client information will enter it, because informed consent requires specifics. Opinion 512 asks for particulars about the kinds of client information disclosed and the risks. Put the product names in a short schedule rather than the body, so you can update the list without re-papering every engagement, and tell clients you will notify them of material changes.
How do I add an AI clause for existing clients?
Send a short addendum or a one-page notice, and ask for a signed or emailed acknowledgment. Say what changes (a new tool touches their file), what does not change (your judgment and supervision), and how to opt out. If a tool will receive their confidential information, treat the addendum as a request for specific informed consent, not a notice, and keep the reply in the file.
Can I charge clients for the AI tools I use?
It depends on the cost. Under Opinion 512, a tool you use across your practice, such as an AI feature inside your word processor, is overhead. A per-use charge from a third-party AI service incurred for one client's matter can ordinarily be billed at actual out-of-pocket cost. California's 2026 guidance adds that no markup may be added without the client's informed written consent. Disclose the approach in the fee agreement.
Does an engagement letter clause protect privilege if my client uses a chatbot?
No clause can make a communication privileged. It can warn the client before the exposure happens. In United States v. Heppner (S.D.N.Y. 2026), the client alone, without counsel's direction, fed information from his lawyers into a consumer AI tool, and the court held the resulting documents neither privileged nor work product. A written warning shows you told him.
Keep the review promise the clause makes
Tiers A and B tell the client a lawyer reviews AI-assisted work, and Tier B says every authority is checked against its source. One cheap safeguard before a draft leaves the firm is the free Hallucination Shield. It checks up to 50 citations per run, needs no signup, and flags any citation it cannot verify. Whether a holding was described correctly is still your read.
This article is for general information and is not legal advice. For guidance on a specific case, consult a licensed attorney in your jurisdiction.