Caution: Potential Privilege Waiver Pitfalls in AI Use by Clients

Lawyers and clients are increasingly using artificial intelligence ("AI") and finding incredible efficiencies. But two separate decisions from federal courts in February demonstrate that clients should be cautious about inputting attorney communications or work product into public AI platforms because doing so might waive any privilege or confidentiality that would normally attach to such information.

Confidential client communications with an attorney for the purpose of receiving legal advice are privileged. In litigation, an opposing party cannot discover these communications unless privilege is waived. Privilege belongs to the client, so when clients disclose attorney-client communications to a third party, they typically waive that privilege. Separately, attorney work product is information generated for litigation either by an attorney or at an attorney's direction, which could include documents created by a client. Work product privilege can be waived if the information is disclosed to a third party or if its confidentiality is not protected.

Clients may want to use AI to organize or develop their thoughts on legal matters and to ask legal questions short of discussing everything with counsel. Potential issues arise when clients upload attorney work product or otherwise privileged communications (e.g., a draft brief or an email) into public AI large language model tools like ChatGPT or Claude. This kind of use raises questions about whether a client's disclosure of attorney communications or work product in public AI tools constitutes disclosure to a third party that waives applicable privileges. It also raises questions about whether clients' use of generative AI platforms creates privileged work product.

Federal courts are developing a framework for analyzing privilege in this context, but the picture is not yet clear. Two cases from February 2026 illustrate this developing case law. First, United States v. Heppner held that a client's communications with public AI platforms, like ChatGPT, Gemini, and Claude, are neither protected by attorney-client privilege nor the work product doctrine. --- F.Supp.3d ---- (S.D.N.Y. Feb. 17, 2026). Second, Warner v. Gilbarco, Inc., held that a client's use of AI tools for drafting assistance constitutes protected work product that opposing parties cannot discover. 2026 WL 373043 (S.D. Mich. Feb. 10, 2026).

The Heppner Decision: Rejection of AI Privilege Protection

The Southern District of New York's decision in United States v. Heppner may be the first federal court decision to comprehensively address privilege issues in AI contexts, describing it as "a question of first impression nationwide." The court rejected both attorney-client privilege and work product protection for a client's communications with Claude, establishing a framework that likely applies to other public AI platforms.

In the criminal proceeding, the defendant had used Claude after receiving a grand jury subpoena to prepare "reports that outlined defense strategy." He input information he had learned through attorney-client communications, created the reports to have more efficient meetings with his attorney, and shared the AI product with his attorney.

The court held that the communications were not between client and attorney because "Claude is not an attorney" and no attorney-client relationship exists with AI platforms. The court rejected arguments that AI platforms function like other software tools, explaining that recognized privileges require "a trusting human relationship" with "a licensed professional who owes fiduciary duties and is subject to discipline," which cannot exist with AI platforms.

For work product protection, the court found that AI-generated documents were not "prepared by or at the behest of counsel" and did not "reflect defense counsel's strategy" at the time of creation. The defendant created the documents "on his own volition," meaning he was not acting as counsel's agent. Ultimately, the court ordered disclosure of his prompts and the AI product to the prosecution.

The Warner Decision: Protection for AI Usage as Work Product

The Eastern District of Michigan's decision in Warner v. Gilbarco, Inc., took a different approach, holding that a litigant's use of AI tools for drafting assistance constitutes work product protected from discovery. The court denied a motion to compel production of information concerning plaintiff's use of third-party AI tools in connection with the lawsuit, finding that such information was not discoverable.

The court characterized AI platforms as "tools, not persons" and explained that work product waiver requires disclosure "to an adversary or in a way likely to get in an adversary's hand." Because AI tools like ChatGPT are not adversaries, and defendants presented no evidence that the plaintiff's use of AI made it likely that defendants would access the protected information, the court found no waiver.

The Warner court's protective approach applied standard work product doctrine to prevent discovery of the plaintiff's litigation preparation methods, including AI usage. The decision does not address attorney-client privilege waiver or situations involving more extensive AI data sharing that might substantially increase adversary access to protected information.

Takeaway: Be Cautious While Case Law Develops

AI is a valuable tool, but clients and attorneys must also understand its risks.

The distinction between public and private AI platforms appears crucial for privilege analysis. Public platforms like ChatGPT, Claude, and others typically involve data sharing that courts may view as third-party disclosure. Enterprise or private AI solutions with stronger confidentiality protections may receive different treatment, though no federal court has yet addressed this distinction directly. Clients and attorneys should be especially cautious when using public AI platforms, and they should have a discussion at the outset of their relationship about the topic to ensure otherwise privileged information is not inadvertently disclosed.

Second, clients should bear in mind the uncertainty in this developing case law. A court might determine that the client's usage of AI—not only the prompts used but also the resulting AI-generated content—is discoverable. It is best practice to consult with your attorney before using generative AI and inform your attorney about your AI usage.

We Can Help

Maslon's attorneys are ready to answer your questions about how to use generative AI while steering clear of potential privilege pitfalls.