What your client tells an AI.

A client used a consumer chatbot on his own case. The court said the conversations were not privileged.

In United States v. Heppner (S.D.N.Y., February 2026), a represented defendant had used a consumer AI service on his own to analyze his case and had shared the results with his lawyers. The court held that the exchanges with the AI were not privileged: the tool is not an attorney or an attorney's agent; a consumer product under consumer terms carries no reasonable expectation of confidentiality; and the client was not using it to obtain legal advice from counsel.

The decision is narrower than its headlines. It turned on a consumer product, consumer terms, and use without counsel's direction, and parts of its reasoning have been questioned since. Two later federal decisions reached different results on different facts, and the first state-court decisions have followed. None of that helps the client who has already typed the story into a free chatbot.

The statute covers the attorney's inputs. The client's are outside it.

Business and Professions Code § 6068.1 tells an attorney where confidential information may go: only into a system whose access is restricted to the attorney and to people bound by confidentiality. It is a duty on the attorney, about the attorney's own use. A client at home, typing the facts of a dispute into whichever assistant came with the phone, is not within it, and after Heppner may not be within the privilege either.

That is now the ordinary case. Clients arrive having already asked an AI what their case is worth, what to say, and what the other side will argue. The output comes to the lawyer as an e-mail. Every one of those conversations is a document that may be asked for later.

What to tell clients, in writing, at the start.

• Keep the case out of public AI. No consumer chatbot, free or personal, for anything about the matter. Say it in the engagement letter, in plain words, with the reason.
• Tell the lawyer, not the machine. A client who needs to think a problem through in writing can write notes for counsel. Notes prepared to communicate with a lawyer have long been treated as part of that communication; a chat with a product under its own terms is a different thing.
• If it already happened, say so. A client who has used an AI tool on the matter should tell the lawyer what was entered and when, so the lawyer can decide what it means before anyone else does.

These are practices many firms are adopting; what any firm should say in its own letters is the firm's judgment on its own facts.

A private place to tell the story, inside the matter the attorney works in.

ClearlyConcise gives the client somewhere else to go. At the attorney's invitation, the client tells the story in their own words, answers the questions it leaves open, and sends the documents they already have, inside the same sealed matter file the attorney works in, under the firm's supervision. The first thing the client reads there is the instruction above: keep your case out of public AI tools. Nothing the client writes reaches anyone outside the matter, and nothing reaches the client that an attorney of the firm did not approve.

What that does for the privilege question is for the firm and its counsel to judge; the workspace creates no privilege and makes no representation about how a court will read it. What it does for the practice is plain: the client's account is taken once, in a place built for it, instead of in a chatbot's history.

Where the law stands, in one list.

• United States v. Heppner, No. 25-cr-00503 (S.D.N.Y. Feb. 17, 2026): a represented client's exchanges with a consumer AI tool held not privileged.
• Warner v. Gilbarco, Inc., No. 2:24-cv-12333 (E.D. Mich. Feb. 10, 2026): a motion to compel all AI-use materials denied; the court treated the tools as tools, not persons, for work-product purposes.
• Morgan v. V2X, Inc., No. 25-cv-01991 (D. Colo. Mar. 30, 2026): work-product protection extended to AI-assisted preparation on a consumer tool; the identity of the tool held discoverable.
• State courts have begun to rule (Texas and New York, mid-2026), and federal courts have begun approving protective orders that name or describe the AI tools parties may use on produced material.
• California: Senate Bill 574, signed September 30, 2026; § 6068.1 effective January 1, 2027. What it requires →

Trial-level decisions, unreconciled, cited by their docket numbers and dates; read the opinions before relying on any of them. No California court has decided the question as of this writing.

A demonstration, under your firm's letter.

If you practice in California and want to see the client's side of the workspace on a matter in your practice area, write to me. Forty-five minutes, synthetic facts, a short confidentiality letter. First matters begin after the independent security test reports in November.

johnjensen@johnmjensen.com

This page is general information about reported decisions and a statute. It is not legal advice, and reading it does not create an attorney-client relationship. John M. Jensen is the founder of ClearlyConcise, Inc.; the views are his own.