Artificial intelligence has quietly become the newest member of many project teams. Estimators use it to check numbers, project managers use it to draft correspondence, and executives use it to think through alternate perspectives on thorny problems. It feels private, like you are carrying on a conversation with a tireless assistant. But a wave of recent court rulings delivers an uncomfortable reminder: what you type into an AI tool may not be protected if your dispute lands in court.
Two protections are at stake. The attorney-client privilege shields confidential communications with your lawyer. The work-product doctrine shields materials prepared in anticipation of litigation. Neither was written with chatbots in mind, and courts are now sorting out how — and whether — they apply.
The leading cautionary tale is United States v. Heppner, where a federal judge in the Southern District of New York held that a defendant’s chats with a popular AI tool were neither privileged nor protected work product. The court reasoned that, because the AI model is not a lawyer, privilege never attached to the conversation with the model. Further, because the tool’s own privacy policy clearly stated that inputs and outputs could be stored, used to train the model, and shared with third parties, the “conversation” was really a disclosure to an outsider, waiving confidentiality. Forwarding the results of an AI conversation to counsel afterward could not retroactively cloak the materials in privilege.
Other courts have been friendlier, though only moderately so. In Warner v. Gilbarco and Morgan v. V2X, federal courts protected self-represented parties’ AI work as work product, reasoning that a chatbot is a tool, not an adversary, so feeding it your thoughts is not the kind of disclosure that forfeits protection. A New York state court reached the same result this June in Assini v. Hayward, quashing a subpoena for a party’s chatbot prompts and outputs. But even these decisions came with strings attached: the Morgan court still ordered the party to reveal which AI tool he used and barred him from uploading confidential information into mainstream platforms, and the Assini court warned that careless AI use could invite sanctions.
The through-line is a distinction most clients already know well: “open” versus “closed.” Public, consumer-grade tools, like free chatbots, are the riskiest. Enterprise or “closed” systems, used under contracts that forbid training on your data and limit disclosure, look much more like the confidential, supervised setting these protections were built to guard. As courts continue to assess this issue, that line will likely develop further nuances.
Cases discussed: United States v. Heppner (S.D.N.Y. Feb. 17, 2026); Warner v. Gilbarco, Inc. (E.D. Mich. Feb. 10, 2026); Morgan v. V2X, Inc. (D. Colo. Mar. 30, 2026); Assini v. Hayward, 2026 NY Slip Op 26086 (Sup. Ct. Nassau County June 4, 2026).
What this means for you
Based on limited case law to date, using a public AI chatbot can be like shouting your secrets in a crowded elevator. Treat it that way, not as a meeting behind closed doors with legal counsel, and assume that anything you type into a free tool could someday surface in discovery. If you want AI in the mix on a matter that may become a dispute, involve legal counsel early. Work performed by your lawyer or at your lawyer’s direction, inside a vetted closed system with contractual protections and model training switched off, stands the best chance of staying privileged. And read the terms of service before you hit enter: the fine print is the ballgame.
This article is for educational purposes only and is not legal advice. Seek the services of competent legal counsel to address your specific situation or issue.