AI, Confidentiality, and the Firm's Money: What ABA Opinion 512 Actually Requires

ABA Formal Opinion 512, issued in July 2024, is the first national ethics framework for lawyers using generative AI — and its core requirement is simpler than the anxiety around it suggests: use AI competently, protect client confidentiality, and supervise what the tool produces. What it doesn’t do is tell you which software to trust. That judgment stays with the firm. This post translates the opinion into the decisions a managing partner or firm administrator actually has to make, including the one most guidance skips: how the rules apply to the firm’s financial data, not just its case files.

The three duties Opinion 512 turns on

The opinion doesn’t invent new obligations. It maps existing Model Rules onto a new tool:

• Competence (Rule 1.1). You have to understand, at a working level, what an AI tool does and where it fails. You don’t need to build models, but you can’t treat outputs as authoritative without knowing they can be wrong. This is why verifying AI-generated work matters — our AI citation verification checklist exists precisely because confident-sounding output can be fabricated.
• Confidentiality (Rule 1.6). You must protect all information relating to a client’s representation. With AI, the exposure point is the input: what you put into the tool, and what the tool is permitted to do with it.
• Supervision (Rule 5.3). You’re responsible for the AI’s output the way you’re responsible for a non-lawyer assistant’s work. “The model said so” is not a defense.

Read together, these three establish a standard that’s demanding but achievable: know the tool, control the inputs, check the outputs.

The part most firms miss: informed consent and the input

Opinion 512 makes clear that a lawyer must be careful about feeding confidential information into a generative AI tool, and that in some circumstances client consent is part of the analysis. The practical takeaway is to focus on the input decision. Before confidential information goes into any tool, the firm should know whether that tool trains on inputs, where the data is stored, and who can access it.

Say the hard thing plainly: pasting client information into a public, consumer AI tool that trains on user data can breach Rule 1.6 on its own — no breach, no leak, no bad actor required. The disclosure is the violation. That’s why governance has to happen at the point of input, before anything is typed, not after.

Where the financial data fits

Most Opinion 512 commentary stops at documents. But a firm’s financial data is client data too. When an operator wants to ask an AI tool “which matters are dragging our realization this quarter?” — that query touches billed amounts, write-offs, and client payment behavior, all of which relate to representations and all of which Rule 1.6 covers.

This is the connection between ethics and Legal Revenue Operations that rarely gets drawn. The same discipline that protects a client’s case strategy has to protect the client’s numbers. And that’s far easier when the numbers live in one governed system than when they’re scattered across spreadsheets, exports, and email attachments — because you can only supervise access to data you can actually see. Financial data inside LeanLaw is permissioned by role: the firm decides who can view and query revenue performance, and that control holds whether the data is being read in a report or asked about in plain language. That’s a concrete answer to Rule 5.3’s supervision requirement, not a slogan about security.

Turning the opinion into firm policy

Opinion 512 sets the standard; your policy operationalizes it. A defensible policy does four things:

1. Names approved tools — and just as importantly, names the tools that are off-limits for confidential data.
2. Classifies data — what’s safe for general AI use, and what only goes into tools that meet the firm’s confidentiality bar.
3. Assigns supervision — who reviews AI output before it reaches a client or a court.
4. Addresses the financial data — who can use AI on the firm’s revenue and client-financial data, and under what controls.

You don’t need a fifty-page document. A one-page policy that people actually read beats a comprehensive one they ignore. Our AI policy template gives you a structure to adapt, and our guide to the data-privacy implications of AI tools covers the confidentiality analysis in more depth.

The firms that handle this well won’t be the ones that banned AI or the ones that adopted it blindly. They’ll be the ones that read the rule, made deliberate choices about their data, and built those choices into the systems where the data already lives.

Frequently asked questions

What is ABA Formal Opinion 512? It’s the American Bar Association’s first formal ethics guidance (July 2024) on lawyers’ use of generative AI. It applies existing Model Rules — competence, confidentiality, and supervision — to AI use rather than creating new rules.

Does Opinion 512 ban any AI tools? No. It doesn’t endorse or prohibit specific products. It sets the ethical standard and leaves the tool-selection judgment to the firm.

Do I need client consent to use AI on their matter? It depends on the tool and the data. Opinion 512 indicates that informed consent can be part of the analysis when confidential information is involved, particularly with tools that don’t adequately protect inputs. Using a tool that doesn’t train on your data and isolates it reduces — though doesn’t automatically eliminate — the consent question.

How does Opinion 512 apply to billing and financial data? The duty of confidentiality under Rule 1.6 covers all information relating to a client’s representation, including financial details. AI use touching billing, trust, or payment data is subject to the same confidentiality and supervision duties as AI use touching case files.