ABA Opinion 512 article cover

In July 2024, the ABA issued Formal Opinion 512 on generative AI and the duty of competence. Most articles about it either bury you in ethics citations or reduce it to "just be careful." This piece is for managing partners and operations leads who need to actually do something — by Monday.

The opinion is short — 16 pages. But its implications for law firms evaluating AI tools are significant. At the core: ABA Model Rule 1.1 (competence) now explicitly includes understanding generative AI tools you use, or choose not to use. Ignorance is no longer a defense.

What Opinion 512 actually covers

The opinion addresses four core duties from the Model Rules of Professional Conduct: competence (Rule 1.1), confidentiality (Rule 1.6), communication with clients (Rule 1.4), and supervision (Rules 5.1 and 5.3). Each has direct implications for how a firm selects, configures, and monitors AI tools.

The duty that catches most firms off-guard is supervision. Under Rule 5.3, you're responsible for the AI's output the same way you're responsible for work done by an associate. If the AI hallucinates a case citation and it ends up in a brief, that's your firm's problem — not the vendor's.

The opinion states that lawyers must understand "the benefits and risks associated with relevant technology." For AI tools, this means understanding the data handling model, the training data, and the output review process — not just using the tool and hoping for the best.

The 5 things your firm needs to verify

Before deploying any AI tool in your practice, these are the five questions Opinion 512 effectively requires you to answer. We've mapped each to the relevant rule section so you can show your work to the bar if it ever comes up.

1. Is client data used to train the model?

Under Rule 1.6, confidentiality applies to information relating to the representation. If the vendor uses your firm's queries or documents to improve their model, that's a potential 1.6 violation. Require a written confirmation that data is not used for training — and put it in the DPA, not just in a sales email.

2. Is your data isolated from other clients of the vendor?

Multi-tenant architectures where your documents share infrastructure with other law firms create privilege contamination risk. The opinion doesn't say "use single-tenant infrastructure" explicitly, but the supervision duty effectively requires you to verify, in writing, the isolation architecture.

3. Does the tool cite its sources?

Opinion 512 implicitly requires that AI output be verifiable. A tool that generates text without citations to source documents cannot be effectively supervised under Rule 5.3. Every AI output that ends up in a client matter should be traceable to a specific document. If your tool doesn't cite, you can't supervise.

4. Have you updated your engagement letters?

Rule 1.4 (communication) requires that clients be informed about significant decisions affecting their representation. Several state bars have issued guidance that AI use in client matters should be disclosed. Update your engagement letter template now — the language doesn't need to be alarming, but it needs to exist.

"Have you updated your engagement letters?" is the question that flushes out the firms that have actually read the opinion from the firms that read a summary on LinkedIn.

5. Is there a written supervision policy?

Rule 5.3 requires reasonable efforts to ensure non-lawyer assistance conforms to professional obligations. AI is non-lawyer assistance. Your firm needs a written policy on how AI output is reviewed before it reaches a client or a court filing. If a junior associate prepared a draft, you'd review it. AI is no different.

State bar variations matter

Opinion 512 is the ABA's position — but state bars can and do issue their own guidance. California, New York, and Florida have all issued state-specific guidance that goes further than the ABA opinion on certain points. If your firm practices in multiple states, the strictest applicable guidance controls.

The practical upshot: don't just satisfy Opinion 512. Satisfy the strictest state bar guidance you're subject to. For most firms, that's California's December 2023 guidance, which is materially stricter than the ABA opinion on disclosure requirements.

Full disclosure: I co-founded DVLP Studio and we build a product called Legal Intelligence. It's designed around the five verification questions above. Per-firm data isolation. No model training on your data, written into the DPA. Every answer cites the specific source document. We provide engagement letter language and a sworn declaration template for outside counsel guidelines. A written supervision policy for your firm to adapt.

You don't need our product to satisfy Opinion 512. But you do need to verify, in writing, that whatever AI tool you're using satisfies these five questions. The opinion makes it your obligation, not the vendor's.

If you want to talk through how your firm is approaching this, get on a call. I'll be direct about whether Legal Intelligence is the right fit — and tell you who else to look at if it's not.