Does Using AI Waive Attorney-Client Privilege? Confidentiality When Lawyers Use AI

Short answer: Does using AI waive attorney-client privilege? No rule we found says so on its own, but a reported federal ruling, United States v. Heppner, held that a client's own chats with a consumer AI tool were not privileged, in part because the tool's privacy policy allowed disclosure. What you type into an AI tool may be kept, read or produced by the provider, depending on its terms, and ABA Formal Opinion 512 treats that as a Rule 1.6 confidentiality question.

What does the confidentiality rule require before I use AI?

You must weigh the risk that information you put into a tool could be seen by people outside the firm, or by people inside it behind an ethical wall. ABA Formal Opinion 512 says lawyers must check the provider's terms to "learn who has access to the information that the lawyer inputs into the tool" (p. 7).

The opinion also warns that "self-learning" tools can expose one client's information to other users of the same tool (p. 7).

What should I check about how an AI tool keeps my data?

Opinion 512's vendor checklist (p. 11) tells lawyers to determine whether the tool "retains information submitted by the lawyer before and after the discontinuation of services or asserts proprietary rights to the information."

It also asks lawyers to make sure they "will be notified in the event of a breach or service of process regarding production of client information" (p. 11). In practice, that means reading the Terms of Use and privacy policy, which the opinion says "all lawyers should read and understand" (p. 7).

For self-learning tools, Opinion 512 says yes: "a client's informed consent is required prior to inputting information relating to the representation into such a GAI tool" (p. 7). A general clause in the engagement letter is "not sufficient" (p. 7).

States differ. California says not to input confidential information into a tool that "may present material risks to confidentiality or security, absent informed client consent" (2026 Practical Guidance). Texas Opinion 705 asks for consultation and consent only if you are not reasonably sure the tool keeps information confidential. See State rules for others.

No consent is needed when no client information goes into the tool, for example when brainstorming (Opinion 512, p. 7).

Are a client's own chats with a consumer AI tool privileged?

A federal judge has held that a client's own chats with a consumer AI tool were not privileged. In United States v. Heppner, No. 25 Cr. 503 (S.D.N.Y.), Judge Jed S. Rakoff ruled from the bench on February 10, 2026 and issued a written memorandum on February 17, 2026.

According to Proskauer, O'Melveny, Venable and the Harvard Law Review blog, the court held that a criminal defendant's own conversations with the consumer version of Claude were protected by neither attorney-client privilege nor work product. The reported reasons:

  • Claude is not a lawyer.
  • The provider's privacy policy, which permits disclosure including to government authorities, defeated any reasonable expectation of confidentiality.
  • The defendant did not act at his lawyer's direction.

Our account comes from these secondary sources, not from the memorandum itself.

Does Heppner apply to enterprise AI tools or AI used at a lawyer's direction?

The ruling, as reported, did not decide either question. It is one trial-court decision, and commentators, including the Harvard Law Review blog and Lawfare, dispute its reasoning.

Opinion 512 does not address privilege waiver directly; its focus is the Rule 1.6 duty of confidentiality (our reading).

Can a court order an AI company to keep chats users deleted?

Yes, it has happened. In In re OpenAI, Inc. Copyright Infringement Litigation, No. 23 Civ. 11195 (S.D.N.Y.), the lawsuit brought by The New York Times, Magistrate Judge Ona T. Wang entered a preservation order against OpenAI on May 13, 2025 (ECF No. 551). Duane Morris summarizes the order as requiring OpenAI "to preserve and segregate output log data that would otherwise be deleted, including data subject to user deletion requests or statutory erasure rights" (see also OpenAI). That included chats users had deleted and data subject to statutory erasure rights.

Which accounts did the OpenAI preservation order cover, and which did it exempt?

According to OpenAI's own statement, the order covered ChatGPT Free, Plus, Pro and Team, and the OpenAI API without a Zero Data Retention agreement.

OpenAI says the order "does not impact ChatGPT Enterprise or ChatGPT Edu customers," nor "API customers who are using Zero Data Retention endpoints," and that on May 27, 2025 "the Court clarified that ChatGPT Enterprise is excluded from preservation." These exemptions are OpenAI's description; OpenAI is a party to the case.

Is the OpenAI preservation order still in effect?

The ongoing duty has ended. A stipulated order on October 9, 2025 ended it as of September 26, 2025, kept the logs already segregated, excluded data from the EEA, Switzerland and the UK, and added targeted preservation for flagged accounts (Duane Morris).

OpenAI's October 22, 2025 update says: "Our obligations under the earlier order ended on September 26, 2025."

What does this mean for lawyers?

Our analysis: a provider's promise to delete your data can be overridden by a court in someone else's lawsuit, and the terms of your plan can decide whether a judge sees a reasonable expectation of confidentiality. Read the terms of the exact tier you use, and know whether it keeps your prompts and files.

For a plan-by-plan answer on ChatGPT, see Is it safe for a lawyer to put client files in ChatGPT?


Last reviewed November 16, 2026. This is general information, not legal advice.

About Faradex. Faradex is a private AI workspace for small and mid-sized firms that handle confidential client work. Each firm gets its own Workspace on dedicated infrastructure in the United States. Files and conversations are held in memory and are gone when the Workspace is reset. AI model: Currently Opus via AWS Bedrock, more information is available on our website. AWS does not store prompts or responses, and model providers have no access to them. Pricing: from $120 a month, no seat minimum; see faradex.ai/pricing.