ABA Opinion 512: what it actually means for your firm

What ABA Opinion 512 actually says

Formal Opinion 512 is the ABA's first comprehensive guidance on generative AI in legal practice. It does not invent new duties. It walks through Rules 1.1, 1.6, 5.1, 5.3, and 1.5 and explains how each one applies when a lawyer or a non-lawyer staff member uses a generative AI tool on a client matter.

The Opinion is persuasive rather than binding. State bars have started to align with it and most large firms now treat 512 as the floor, regardless of jurisdiction, because state regulators tend to follow the ABA over time and clients have started asking for it in vendor reviews.

Competence does not mean reading the model paper

Rule 1.1 requires lawyers to understand the benefits and risks of the technology they use. For AI, that means knowing what a tool can and cannot do, where its outputs come from, and how to verify them. It does not mean every partner needs a working theory of transformer architectures.

The practical test is whether a lawyer can describe, in plain terms, what the tool was trained on, what data the firm sends to it, what comes back, and how the lawyer checked the output before relying on it. If the answer to any of those is vague, the competence duty is not met.

Confidentiality is about the data, not the brand

Before pasting client material into a tool, a lawyer needs informed consent or a clear contractual basis that the material will not be used to train models, exposed to other tenants, or retained beyond the engagement. Public chatbots that train on user input are usually not appropriate for client material. Enterprise tools with a no-training contract usually are.

The line that matters is the data path, not the brand on the login screen. The same vendor often offers a consumer product trained on inputs and an enterprise product that is not. The Opinion expects firms to know which one their lawyers are actually signed into.

Supervision applies to associates and non-lawyer staff alike

Rules 5.1 and 5.3 put partners on the hook for the AI use of associates and non-lawyer staff. The Opinion expects reasonable supervisory measures: a written policy, training, an approved-tools list, and a way to check that the policy is being followed in practice.

Telling associates not to do something once at orientation does not satisfy the duty. Neither does a policy that no one can find. Supervision is a system, not a memo, and the Opinion makes clear that the system itself is what is being inspected.

Fees: AI-assisted time is not human time

Rule 1.5 is unchanged. Time saved by an AI tool cannot be billed as if a human did the work, and bills that obscure AI-assisted time risk being unreasonable. Most firms are updating their engagement letters to address this explicitly, both to inform the client and to lock in the firm's billing posture before a disputed invoice surfaces it.

What to put in place this quarter

• An approved-tools list with a documented review process and an owner
• A one-page firm AI policy with examples written for associates
• Short training plus a signed acknowledgement for every user with AI access
• Engagement-letter language covering AI use and billing
• A quarterly compliance review with named participants and a written agenda

Common mistakes firms make on 512

• Treating 512 as IT's problem instead of the managing partner's
• Approving a tool because it is enterprise-branded without reading the data terms
• Writing a policy that does not name specific tools, leaving associates to guess
• Skipping training for non-lawyer staff even though Rule 5.3 covers them
• Treating the engagement letter as static after the first AI matter ships

Build a 512-aligned program with CounselGuard

CounselGuard maps every Rule under 512 to the evidence your firm can produce, from approved-tools registry and policy approvals to per-matter activity capture and quarterly audit reports. The same workflow that satisfies 512 satisfies the FLSC, SRA, and EU AI Act overlaps.