Almost everything in this section is about building things. This one is about what you owe the person on the other end of them.
Two questions, and they get very different treatment. The first — must you tell the client — is genuinely unsettled and reasonable lawyers disagree. The second — what happens to the bill — is not unsettled at all, and the answer is the one lawyers least want to hear.
The fee question, which is settled
Start here, because it is clear and because it is where the money is.
The ABA’s Formal Opinion 512, issued July 29, 2024, addresses generative AI under the familiar rules, and its treatment of fees under Model Rule 1.5 makes two points that leave little room.
You may not bill a client for time spent learning a technology for general use in your practice. Learning your tools is overhead. It is the same principle that prevents billing a client for learning to use a new research database or a new document management system. (The opinion treats it differently where a client specifically requests the use of a particular tool on their matter.)
And if the work took less time, you bill less time. A task that used to take two hours and now takes twenty minutes is a twenty-minute entry. The efficiency belongs to the client.
That second one is worth sitting with, because it is the load-bearing sentence of this entire section. Reasonable-fee rules have always keyed to time actually and reasonably spent. When a tool compresses the work, the entry compresses with it. There is no version of the rules under which the lawyer captures the gain by continuing to bill the old number.
Which produces an uncomfortable arithmetic for hourly practice: the better you get at this, the less you bill for the same work. That is not a defect in the rules. It is the rules functioning exactly as designed, and it is the single strongest argument for moving compressible work to fixed fees — which lets a firm keep the efficiency it created without billing for time it did not spend, and which gives the client a number in advance.
The dishonest resolution — same hours, less work — is the one that would quietly negate everything this section is for. If AI-assisted practice just means the same bill for less effort, no one outside the firm is better off, and the access argument collapses. The whole reason to publish any of this is the premise that the gain shows up as a smaller bill or as a case somebody could otherwise not afford to bring.
The disclosure question, which is not
Now the harder one. Must you tell the client?
Rule 1.4 requires a lawyer to “reasonably consult with the client about the means by which the client’s objectives are to be accomplished,” to keep the client reasonably informed, and — at subsection (b) — to “explain a matter to the extent reasonably necessary to permit the client to make informed decisions regarding the representation.”
There is no rule that says “disclose AI use,” and I do not expect one that says it in those words, for the reason I have set out elsewhere: within a few years that will describe every document produced by every lawyer, the way “used a computer” does now. A disclosure requirement that captures everything communicates nothing.
But 1.4 is not a checklist, and the honest answer is that it depends on facts. Here is how I sort it.
Disclosure is clearly required where client data leaves your control in a way the client would care about. If a client’s confidential information is being transmitted to a third-party service, that is a fact about the representation the client is entitled to know and, depending on the arrangement, to consent to. This is the strongest case and it is not close. Formal Opinion 512’s confidentiality analysis runs through the duty to protect information relating to the representation absent informed consent, and a client cannot consent to something they have not been told.
Disclosure is clearly required where the client asked. Obviously. And clients are asking more often.
Disclosure is required where it bears on a decision the client has to make. If the choice of approach changes the cost, the timeline, or the risk profile in a way that matters, 1.4(b) reaches it — the client cannot make an informed decision about a thing they do not know about.
Disclosure is probably not required for ordinary internal tooling that never transmits their information. A local model classifying incoming mail is a piece of office equipment. We do not disclose our word processor, our research database, or our calendaring system, and the analytical work is being done by “does the client’s information go anywhere” rather than by the label on the tool.
And there is a category I would call prudent regardless of whether it is required: institutional clients with their own AI policies, matters where the work product is unusually sensitive, and any engagement where you would be uncomfortable if the client learned about it later from someone else. That last test has served me well across a lot of disclosure questions that had nothing to do with technology.
The engagement letter is the right place
The practical resolution to most of the above is not a conversation. It is a paragraph.
An engagement letter that states plainly what technology the firm uses, what happens to the client’s information, and what remains under lawyer supervision does the 1.4 work up front, in writing, at the moment when the client is actually deciding whether to engage you. It converts a recurring judgment call into a settled term.
What such a paragraph should cover, in plain language rather than in the language of a software license:
- That the firm uses technology including AI tools to work more efficiently.
- Whether client information is transmitted outside the firm and, if so, under what confidentiality terms.
- That all work product is reviewed by a lawyer, who is responsible for it.
- That the client may ask questions about it at any time.
I would resist the temptation to make this long. A dense technical disclosure that nobody reads is worse than a short one that somebody does — it produces the appearance of consent without the substance, which is the thing informed-consent doctrine exists to prevent.
And a caution about the opposite failure: a clause broad enough to authorize anything is not informed consent either. “Client consents to the firm’s use of technology” tells the client nothing and would not survive a serious challenge. If the point is that they can make an informed decision, the paragraph has to contain information.
What clients actually ask
An observation rather than a rule, from having had the conversation a number of times now.
Clients rarely ask whether you use AI. They ask two other things:
“Is a real lawyer looking at this?” They want to know that somebody is answerable. This is the right question and it has a clean answer: yes, and here is what that means. Nothing goes out unreviewed, a lawyer signs, and the responsibility for the work is unchanged.
“Am I paying less?” Also the right question, and the one that determines whether they believe the first answer. A client who is told their lawyer is dramatically more efficient and then receives the same bill draws an obvious and correct conclusion.
Which brings the two halves of this piece together. The fee answer is the disclosure answer, in practice. A firm that passes efficiency through has an easy conversation about technology, because the client can see the benefit. A firm that does not will find the disclosure conversation increasingly uncomfortable, and should.
The line that does not move
One thing is not a disclosure question, a fee question, or a policy question, and it is worth stating flatly.
The lawyer is responsible for the work. Not “responsible for supervising the tool.” Responsible for the work, in the same way and to the same extent as if every word had been typed by hand at two in the morning. A brief with a bad citation in it is your bad citation. A missed deadline is your missed deadline. An error in a document you signed is yours.
That is not going to change, and I would not want it to, because it is the thing that makes any of the rest of it safe to do. Everything in this section — the verification gate, the specialist chain, the refusal to guess, the human confirmation points — exists because someone has to answer for the output, and that someone is a person with a license.
What you owe the client you automated is exactly what you owed them before: competent work, honest billing, enough information to make their own decisions, and a name on the document.
Sources
- ABA Standing Committee on Ethics and Professional Responsibility, Formal Opinion 512 (July 29, 2024)
- Minn. R. Prof. Conduct 1.4 (Communication), Rule 1.5 (Fees), Rule 1.6 (Confidentiality)
Rules quoted were verified against the Revisor of Statutes. This is general commentary, not legal advice and not ethics advice. Formal Opinion 512 is advisory and not binding in any jurisdiction; disclosure obligations depend on your jurisdiction, your engagement terms, and the facts of the representation.