What ABA Formal Opinion 512 means for a small firm using AI

In July 2024 the American Bar Association's Standing Committee on Ethics and Professional Responsibility issued Formal Opinion 512, "Generative Artificial Intelligence Tools." It was the ABA's first formal opinion on the subject, and it didn't create new rules. Instead, it walked through how the Model Rules you already follow apply when a lawyer uses tools that draft, summarize or answer questions.

If you run a firm of a few lawyers, you don't need a committee to respond to it. You need to understand the handful of duties it focuses on, and put some simple habits in place.

One caution first: ABA opinions are guidance, not binding law. Your state's rules of professional conduct, and any opinions from your state bar, are what govern you. Several state bars have published their own AI guidance, and they don't all say exactly the same thing. Read yours.

Competence: understand the tool well enough to use it safely

Rule 1.1 asks lawyers to be competent, and since 2012 the comments to that rule have included keeping up with the benefits and risks of relevant technology. Opinion 512 applies that to AI: you don't have to be an engineer, but you need a reasonable understanding of what a tool can and can't do.

In practice, that means knowing:

• What the tool is reading from. Is it working from your file, or from whatever it learned in training?
• How it can be wrong. AI tools can produce confident, fluent text that is simply false.
• How much checking the task needs. The opinion makes the point that the right amount of review depends on the task. A first draft of an internal email needs less than a fact in a demand letter or a citation in a brief.

Confidentiality: know where client information goes

Rule 1.6 requires you to protect information relating to the representation. Opinion 512 spends a lot of time here. Before putting client information into an AI tool, you need to understand whether that information could be retained, seen by others or used to train the tool, and what the vendor's terms actually say.

The opinion draws a distinction worth remembering. Where a tool can learn from what you put in and disclose it outside the firm, the opinion says the client's informed consent is needed, and it specifically says that boilerplate language in an engagement letter isn't enough.

Questions to ask any vendor:

1. Is our data used to train any model, yours or a provider's?
2. Does the AI provider keep our data after it responds? For how long?
3. Who at your company can see our data, and is that access logged?
4. Where is the data stored, and is it kept separate from other customers' data?

If the answers are vague, treat that as an answer.

Communication: when to tell clients

Rule 1.4 covers keeping clients reasonably informed. Opinion 512 doesn't say you must disclose every use of AI. It does say there are situations where you should, for example when a client asks, when the engagement terms require it, or when AI use is relevant to a decision the client has to make. Whether to tell clients more as a matter of practice is a judgment call for your firm.

Candor and meritorious claims: check what you file

Rules 3.1, 3.3 and 8.4(c) mean you can't put false statements of law or fact in front of a court. The well-known cautionary example is Mata v. Avianca, where lawyers filed a brief citing cases that didn't exist. The opinion is clear that relying on an AI tool doesn't change your responsibility for what you sign. We wrote a separate guide to checking citations.

Supervision: set firm rules, even small ones

Rules 5.1 and 5.3 make managing lawyers responsible for reasonable measures so that lawyers and staff follow the rules, including when they use technology. For a small firm, that can be a one-page policy:

• Which AI tools are approved for client work, and which aren't.
• What must always be checked by a lawyer before it leaves the office.
• Who to ask when unsure.
• A short training session when someone joins.

Fees: bill for the work, not the shortcut

Rule 1.5 requires reasonable fees. Opinion 512 addresses a few practical points. If you bill hourly, you bill for the time you actually spent, not the time a task used to take. It also says you generally shouldn't bill a client for time spent learning a tool you'll use across your practice, and that the cost of tools that are part of your general overhead isn't usually something to pass on as an expense. For contingency practices, the broader point still applies: the fee has to be reasonable for the work done.

A one-month plan for a small firm

1. Week one: list every AI tool anyone at the firm uses, including free ones on personal accounts.
2. Week two: ask each vendor the four confidentiality questions above. Stop using any tool whose answers you can't accept for client information.
3. Week three: write the one-page policy. Decide what always needs a lawyer's review.
4. Week four: walk the team through it, and put a date in the calendar to review it in six months.

How we built Meruit around this

We designed Meruit with these duties in mind. Every fact the AI suggests is tied to a source; chronology entries point to a page and a quoted line that Meruit checks is really there. Nothing that leaves the office goes without a named lawyer's approval, and approvals are recorded in an audit log that can't be edited. Firm data isn't used to train models, and we only use AI providers that agree not to keep it. None of that does your thinking for you, and it shouldn't. Opinion 512's central message is that the lawyer stays responsible. Good tools should make that easier, not blur it.