The Risk Radar: AI and Attorney-Client Privilege – Is There a Duty to Warn?

“The implications of AI for the law are only beginning to be explored.” United States v. Heppner, 820 F. Supp. 3d 292, 294 (S.D.N.Y. 2026).

For decades, lawyers have routinely cautioned clients not to discuss their cases with family members, friends, coworkers, or strangers. The reason is apparent: voluntary disclosure of confidential communications to a third party may destroy attorney-client privilege. Now, lawyers may need to add a new warning to that list: do not discuss your case with artificial intelligence.

The rapid adoption of publicly accessible generative AI platforms such as ChatGPT, Claude, and Gemini has created new questions concerning privilege, confidentiality, and discoverability. Clients increasingly use these tools to analyze legal issues, summarize documents, evaluate litigation strategies, and even draft communications to counsel. What many clients do not understand, however, is that information shared with a public-facing AI platform may not receive the same legal protections as information shared with counsel.

That risk became clear in United States v. Heppner, 820 F. Supp. 3d 292 (S.D.N.Y. 2026). After becoming the target of a federal criminal investigation, Heppner used Anthropic’s Claude platform to prepare reports and analyses relating to the investigation. He later claimed attorney-client privilege and work-product protection over approximately thirty-one documents memorializing those communications. The court rejected both arguments.

Relying heavily on the platform’s privacy policies, the court concluded that Heppner lacked a reasonable expectation of confidentiality because Claude retained user inputs and outputs and reserved the right to share information with third parties. The court also characterized Claude as a “third-party AI platform” and noted that Heppner’s attorneys had not directed him to use the tool. Although Heppner later shared the AI-generated materials with counsel, the court emphasized that non-privileged communications do not become privileged merely because they are subsequently provided to an attorney.

Notably, the court suggested that a different analysis might apply if counsel had directed the client to use the platform. Yet, the opinion leaves unresolved whether such instructions could overcome the court’s broader concern that public AI platforms may fundamentally lack the confidentiality required for privilege.

While Heppner is the most direct authority on the issue, subsequent courts have taken a more cautious approach. In Warner v. Gilbarco, Inc., the court rejected an argument that use of ChatGPT by a pro se Plaintiff waived work-product protection, reasoning that AI systems are “tools, not persons.” 820 F. Supp. 3d 629, 636 (E.D. Mich. 2026). Similarly, the courts in Morgan v. V2X, Inc. 2026 WL 864223 (D. Colo. Mar. 30, 2026) and Assini v. Hayward, 261 N.Y.S.3d 915 (N.Y. Sup. Ct. 2026) declined to adopt a rule that use of public-facing AI platforms automatically destroys legal protections. The result is a split in authority, with some courts viewing public AI platforms as third parties and others treating them as technological tools.

For practicing lawyers, however, the more important question is not how courts will ultimately resolve the privilege question, but whether existing ethical rules require attorneys to warn clients about the risks associated with public AI platforms.

Three duties suggest the answer may be “yes.”

First, the duty of confidentiality under ABA Model Rule 1.6 requires lawyers to make reasonable efforts to prevent unauthorized disclosure of client information. If courts follow Heppner and conclude that AI platforms are equivalent to a third party, then allowing clients to submit confidential facts, legal theories, draft pleadings, or attorney communications into those systems creates a foreseeable risk that confidential information will be disclosed. A lawyer aware of that risk may have an obligation to warn against it.

Second, the duty of competence under ABA Model Rule 1.1 requires lawyers to understand the benefits and risks associated with relevant technology. The Connecticut Supreme Court recognized this concern in a footnote in Tov Realty, LLC v. Suarez, warning that competent use of generative AI requires attorneys to understand the technology’s implications for attorney-client privilege, confidentiality, and work-product protection. 355 Conn. 902, 907 (2026). If attorneys are expected to understand those risks, competence arguably requires communicating them to clients.

Third, the duty to communicate material risks under ABA Model Rule 1.4 requires attorneys to explain material risks sufficiently to permit informed client decisions. Whether a client uploads witness statements, medical records, settlement evaluations, financial records, draft pleadings, or privileged attorney communications into an AI platform, the possibility of losing privilege is a material consideration. A client cannot make an informed decision about AI use without understanding that risk.

While courts remain divided on whether the use of public-facing AI platforms constitutes disclosure to a third party, attorneys are faced with addressing discovery issues pertaining to AI. In some jurisdictions, “form” discovery requests approved by the trial court now contain requests expressly seeking information concerning a party’s use of artificial intelligence, including requests for production of AI-generated work product, prompts, inputs, uploaded materials, and outputs. Whether or not those materials will ultimately be protected by subsequent appellate opinions, their inclusion in discovery requests reflects a developing view that AI interactions may be discoverable. Thus, the discovery risk is already posing a practical concern for attorneys litigating their cases at the trial court level.

Just as attorneys routinely advise clients not to post about their cases on social media, forward privileged emails, or discuss litigation with third parties, they should consider giving similar guidance regarding public AI platforms. Accordingly, lawyers should consider providing clients with a simple warning at the outset of representation:

Do not upload, submit, or discuss confidential facts, attorney communications, draft pleadings, litigation strategy, settlement positions, or other case-related information through public artificial intelligence platforms, including ChatGPT, Claude, Gemini, or similar services, without first consulting counsel. Such disclosures may jeopardize claims of attorney-client privilege, work-product protection, or confidentiality.

No ethics rule explicitly requires this exact warning. Nevertheless, as courts increasingly examine whether communications with public AI platforms constitute third-party disclosures, the combined duties of confidentiality, competence, and communication make the case for such a warning increasingly compelling.

Lawyers have long warned clients not to discuss their cases with third parties. The emerging question is whether lawyers now have a duty to warn clients not to discuss their cases with AI platforms.