What ABA Formal Opinion 512 Requires of Lawyers Using AI

When the American Bar Association issued Formal Opinion 512 on July 29, 2024, it did not announce a new rulebook for artificial intelligence. It said the opposite: the Model Rules of Professional Conduct you already follow govern your use of generative AI, and they govern it now.

That is the central, and easily missed, message of the first ABA formal ethics opinion devoted to generative AI. No carve-out exists for tools that are new, impressive, or fast. Competence, confidentiality, communication, candor, supervision, and fees apply to a brief drafted with a large language model exactly as they apply to one drafted by a junior associate. For a firm of three to thirty lawyers, where the managing partner is often also the technology officer and the supervising attorney, the practical takeaway is that you can adopt these tools, but you own the output. This article walks through the opinion rule by rule, with the primary authorities cited so you can read them yourself.

Competence: the hallucination problem is your problem

Model Rule 1.1 requires competent representation, and Comment 8 ties competence to technology by directing lawyers to keep abreast of the benefits and risks of relevant technology. Opinion 512 reads that comment to require a reasonable understanding of the capabilities and limitations of any generative AI tool a lawyer uses. You need not become an engineer, and the opinion expressly declines to say that AI tools are required for competent representation. But you do need to understand enough to use the tool responsibly, and to update that understanding as the technology changes.

The sharpest application is the well-documented tendency of these systems to fabricate. A generative model predicts plausible text; it does not consult a database of real authority unless it is specifically built to, and even then it can misdescribe what it finds. The opinion warns that uncritical reliance on AI output can itself be a competence failure. The duty is verification: every citation, quotation, and proposition that leaves your office must be checked against the actual source.

The case law makes the stakes concrete. In Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023), Judge P. Kevin Castel sanctioned two attorneys $5,000 under Federal Rule of Civil Procedure 11 in an order dated June 22, 2023, after they filed a brief containing fabricated cases generated by ChatGPT, complete with invented quotations and fake internal citations. The court was careful to say there is nothing inherently improper about using a reliable AI tool, but that reliability requires checking the work.

The pattern did not stop there. In Wadsworth v. Walmart, Inc. (D. Wyo. 2025), U.S. District Judge Kelly H. Rankin sanctioned three attorneys under Rule 11 in February 2025 after a motion cited eight cases that did not exist; the lead attorney was fined $3,000 and had his pro hac vice admission revoked, and two others were each fined $1,000. The fabrications came from a firm's in-house AI platform, which underscores a point partners sometimes miss: a tool marketed for legal work is not a substitute for reading the cases.

Confidentiality: read the terms before you paste

Model Rule 1.6 requires lawyers to make reasonable efforts to prevent unauthorized disclosure of information relating to the representation. Opinion 512 applies this to the question every firm faces the first time someone opens a chatbot: can I paste the client's facts in here?

The answer turns on what the tool does with the input. A self-contained or enterprise tool that does not train on your data, retain it for the vendor's purposes, or expose it to third parties presents a manageable confidentiality posture. A consumer tool whose terms allow the provider to retain inputs and use them to train future models, what the opinion calls a self-learning tool, is a different matter. For those, Opinion 512 concludes that a lawyer must obtain the client's informed consent before inputting information relating to the representation. This is the most consequential single holding in the opinion for day-to-day practice, and the heart of any honest conversation about generative AI confidentiality.

Informed consent here is not a line buried in the engagement letter. The opinion states that boilerplate consent will not suffice. It contemplates a real explanation: why the tool is being used, the specific kinds of client information that would be disclosed, the risks including how others might use that information against the client, and the benefits to the representation. The threshold duty, before consent ever arises, is to read and understand the tool's terms of service, data-retention policy, and security practices. You cannot disclose a risk you have not bothered to learn.

A practical confidentiality checklist:

• Confirm whether the tool trains on or retains your inputs, and for how long.
• Prefer enterprise or zero-retention configurations for anything touching client matters.
• Obtain specific informed consent before putting confidential information into a self-learning tool.
• Treat prompts the way you treat email: assume they may be stored, and never paste what you would not send unencrypted.

State guidance points the same way. The Florida Bar issued Ethics Opinion 24-1 on January 19, 2024, advising that protecting client confidentiality is the lawyer's first responsibility when using generative AI and that lawyers should research a tool's data-retention, data-sharing, and self-learning policies before use.

Communication: when you have to tell the client

Model Rule 1.4 governs keeping clients reasonably informed. Opinion 512 does not require a disclosure for every use of AI, but it identifies situations where communication is mandatory. You must disclose when the client asks. You must address it when you are inputting the client's confidential information into a self-learning tool, because that triggers the informed-consent duty under Rule 1.6. And you must disclose when the use of an AI tool is relevant to the basis or reasonableness of your fee.

Beyond those triggers, the opinion treats disclosure as a matter of professional judgment based on the facts of the engagement, and it suggests the engagement agreement as a sensible place to set expectations. The conservative course for a small firm is to address AI use in the engagement letter at intake, so the question is settled before it becomes awkward mid-matter.

Candor and supervision: the duties that catch the whole firm

Two clusters of rules turn an individual lapse into a firm-wide exposure.

Model Rule 3.3 requires candor toward the tribunal, including a duty not to make false statements of law or fact and to correct them when discovered. Opinion 512 connects this directly to AI: a lawyer who files nonexistent opinions or inaccurate analysis generated by a tool has a problem under Rule 3.3, independent of any Rule 11 sanction. The certification you sign when you file is yours, whatever drafted the words.

Model Rules 5.1 and 5.3 extend responsibility upward and outward. Rule 5.1 obligates partners and supervising lawyers to make reasonable efforts to ensure other lawyers in the firm conform to the rules; Rule 5.3 imposes a parallel duty for nonlawyer assistance, which the profession now reads to include AI tools acting in an assistant-like role. Opinion 512 calls on managerial lawyers to establish clear policies regarding the firm's permissible use of generative AI and to train their people on it. In a small firm this is not a corporate-governance exercise. It is a short written policy that says which tools are approved, what may and may not be entered into them, and that all AI-assisted work is verified before it goes out. The supervising partner in Wadsworth was sanctioned alongside the drafter; the order is a reminder that "the associate used the tool" is not a defense.

A minimal supervision policy should cover:

• Approved tools and their data-handling posture, with consumer tools off-limits for confidential input absent consent.
• A non-negotiable rule that every citation and quotation is checked against the primary source.
• Who at the firm answers AI questions and approves new tools.
• Training at onboarding and when tools or rules change.

Fees: bill the time you spent, not the time you saved

Model Rule 1.5 forbids unreasonable fees, and Opinion 512 applies it in a way that disappoints anyone hoping AI is a billing windfall. If a tool lets you produce in one hour what used to take five, you may bill the one hour, not the five. A fee charged for work that was not performed is unreasonable. The efficiency belongs to the client.

The opinion draws two further lines. First, on learning curves: a lawyer generally may not bill a client for the time spent getting up to speed on a tool, because clients should not pay for a lawyer's general education. The opinion recognizes a narrow exception where a client specifically asks the lawyer to use a particular tool for the matter, in which case time learning that specific tool may be chargeable. Second, on whether the tool's cost is a billable expense, the opinion treats it like any other resource: some AI costs are general overhead, absorbed by the firm, while others may be charged to a client if the arrangement is reasonable and disclosed. When AI cost figures into the fee, Rule 1.4 requires telling the client.

What to do Monday

Opinion 512 is permissive in spirit. It does not tell lawyers to avoid generative AI; it tells them to use it the way they are already required to use every other tool, with competence, confidentiality, candor, and honest billing. For a firm of three to thirty attorneys, that translates into a handful of concrete steps: pick tools whose data terms you have actually read, keep confidential information out of self-learning systems unless the client has given specific informed consent, verify every AI-assisted citation before filing, write a one-page firm policy and train on it, and bill the time you spent rather than the time you saved. None of this requires new rules. It requires applying the ones already on the books.

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.