AI Help for Lawyers › Guides › Ethics & confidentiality

Legal AI, ethics, and client confidentiality: the working checklist

Every tool on this site touches privileged material — client documents, matter facts, prospective-client calls. The ethics rules were not suspended for AI, and the bar authorities have now said so explicitly. This is the plain-language checklist; it is not legal advice or an ethics opinion, and your state bar's guidance controls.

Start with Opinion 512 and your state's guidance

ABA Formal Opinion 512 (July 2024) is the anchor document: it applies the existing duties — competence (Rule 1.1), confidentiality (1.6), supervision (5.1/5.3), candor (3.3), reasonable fees (1.5) — to generative AI. The practical upshot: a lawyer must understand, at a working level, how a tool handles client information and how reliable its output is, before using it on client matters. Most states have issued their own guidance layered on top; read yours before the first upload, because the variations (especially on client consent) are real.

The vendor questions that satisfy Rule 1.6

Before any client material reaches a tool, get written answers to five questions. Is our data used to train models — and can that be switched off contractually? How long are inputs and outputs retained, and can we compel deletion? Who at the vendor can access our content, under what controls? Where does processing happen, and is it isolated from other customers? What happens to our data at termination? A vendor that answers crisply is telling you something; so is one that will not. Keep the answers with the engagement file — supervision under Rule 5.3 extends to nonlawyer assistance, and bar guidance treats AI vendors in that frame.

Consent, candor, and the verification duty

Some situations call for informed client consent before AI use — Opinion 512 discusses when self-learning tools and sensitive matters push toward disclosure, and several states go further. Two duties are not situational. Candor: every citation and factual claim an AI produces gets verified by a lawyer before it reaches a court — sanctions for invented citations are now routine enough to need no footnote. Fees: bill AI-assisted work honestly; efficiency gains do not convert into hours not worked.

Make it structural

Write a one-page firm AI policy: approved tools, prohibited uses, the vendor-question file, the citation-verification checklist, and who supervises. Train everyone who touches client matters, including staff. Revisit quarterly — vendor terms and bar guidance are both moving. A policy that lives in a drawer fails Rule 5.3 as surely as no policy; the point is a rhythm the firm actually follows.