What ABA Formal Opinion 512 means for a small law firm

The short answer

ABA Formal Opinion 512, issued on 29 July 2024, is the American Bar Association's first formal ethics opinion on generative AI. It doesn't create new rules. It says the rules you already follow still apply when you use AI: competence, confidentiality, talking to your client, supervising your staff, being straight with the court, and charging a reasonable fee.

For a small firm, it comes down to five habits. Know what your tools do with client data. Get proper informed consent before client information goes into a tool that can learn from it. Check every piece of AI output before anyone relies on it. Make sure everyone in the firm follows the same rules. And bill for the time you actually spend, not the time you would have spent without AI.

Why a California firm should care about an ABA opinion

ABA opinions aren't binding on their own. A San Diego lawyer answers to the California Rules of Professional Conduct and the State Bar of California. But 512 is the most widely cited national starting point, and California is heading the same way.

The State Bar first published its Practical Guidance for the Use of Generative AI in the Practice of Law in November 2023. On 14 May 2026, its Board of Trustees approved an updated version that replaces the 2023 one and covers agentic AI, meaning tools that can start tasks on their own. The State Bar has also put proposed amendments to rules 1.1, 1.4, 1.6 and 3.3 out for public comment. As far as we can see, those amendments aren't in force yet, so check the State Bar's site for where they stand.

What 512 actually asks of you

Competence (Model Rule 1.1). You don't need to be a technologist, but you do need a reasonable grasp of what the tool can and can't do, including how it gets things wrong. Made-up case citations are the famous example. Checking the output is part of the job, and how much checking depends on the task and the tool.

Confidentiality (Model Rule 1.6). Before you put information about a client's matter into an AI tool, you have to weigh the risk that it gets seen by people outside the firm, or by people inside the firm who shouldn't see it. The opinion says that for many self-learning tools, where what you type in can shape what the tool gives other users, you need the client's informed consent first.

Informed consent means a real conversation. The opinion is clear that general boilerplate in an engagement letter isn't enough. The client needs your honest view of why you're using the tool, what client information is involved, the specific risks, how that information could be used against them, and what the tool adds to their matter.

Communication (Model Rule 1.4). Sometimes you'll need to tell the client you're using AI even when you're not putting their information into it, for example when it shapes a decision they'd reasonably want a say in, or when they ask.

Candour and meritorious claims (Model Rules 3.1, 3.3 and 8.4(c)). Anything you file is your responsibility. If AI helped write it, you still check every citation and every statement of fact and law.

Supervision (Model Rules 5.1 and 5.3). Partners and managers need clear firm-wide rules on AI use and need to train lawyers and staff on them. That includes outside vendors who might use AI on your matters.

Fees (Model Rule 1.5). If you bill hourly, you bill for the time you actually spent, even if AI made you faster. With a flat fee, keeping the same fee for work AI let you do in much less time could be unreasonable. You generally can't bill clients for the time it takes to learn a tool you'll use across the firm. The cost of tools that come with your ordinary software is overhead, while a pay-per-use AI service bought for one matter may be billable as an expense if the client knows in advance.

What this looks like in a small firm

The common gap is simple. Someone is already using a personal ChatGPT or Claude account to tidy up a letter or summarise a deposition, and nobody has decided whether that's fine. That's exactly what 512 is pointing at.

A practical starting point:

1. Find out what's already in use. Ask your team which AI tools they use and for what. You'll often find more than you expected. Our post on finding where your data is going walks through how.
2. Pick approved tools and plans. Business plans that don't train on your data and give you admin controls are a very different risk from free personal accounts. Write down which ones are approved.
3. Write a short AI use policy. One page is enough to start: what's allowed, what never goes in a prompt, and who checks the output.
4. Sort out client consent. Decide when you'll ask clients for informed consent and how you'll have that conversation, rather than relying on a line in the engagement letter.
5. Train people and check in. A short session on the firm's rules does more than a long memo nobody reads. Then look again every few months, because the tools and the guidance keep changing.

That's the work our Safe AI Setup is built around: approved tools, a written policy, access controls and staff training. If you'd rather start by seeing where you stand, the AI Safety Check finds out how your team is already using AI. For keeping it right month to month, there's Managed Safe AI.

Frequently asked questions

Is ABA Formal Opinion 512 binding on California lawyers?

No. ABA formal opinions are guidance. California lawyers are bound by the California Rules of Professional Conduct, and the State Bar has its own Practical Guidance on generative AI. In practice the two point the same way, so 512 is a sensible baseline.

Do we need client consent to use AI at all?

Not always. The opinion focuses consent on putting information about a client's matter into tools that can learn from it or expose it to others. Using AI for tasks that involve no client information is a different question. You still have to judge each case, and some clients will want to know either way.

Is a clause in our engagement letter enough?

The opinion says general boilerplate in an engagement letter isn't enough for informed consent. The client needs a real explanation of the tool, the information involved, the risks and the benefits.

Can we bill for time saved by AI?

Not on an hourly basis. You bill for the time you actually spent. The opinion also says you generally can't bill for learning a tool you'll use across your practice, and that ordinary built-in tools are overhead.

What's the first thing a small firm should do?

Find out which AI tools people are already using and on which accounts. Most of the risk 512 describes comes from tools nobody chose on purpose. Once you know, you can pick approved tools, write a short policy and train the team.

*This post is general information, not legal advice. ABA opinions, State Bar guidance and the rules themselves change, so check the current versions and talk to your own ethics counsel about your firm's situation.*