‘AI privilege’ or why and how should our chats with AIs be legally protected?

About:

On June 11, the International Center for Law & Economics, Technische Universität München, and the Data Regulation & Innovation Initiative hosted a moderated discussion on a question rapidly moving from science fiction to subpoena practice: Should conversations with AI assistants be legally protected?

Millions of people now use chatbots to discuss matters they might hesitate to put in an email, say to a co-worker, or admit to a friend. They turn to artificial-intelligence tools to think through legal problems, medical fears, family conflicts, business plans, and private doubts. While these interactions often feel confidential, the law does not yet treat them that way.

The discussion examined recent developments that have made this gap increasingly difficult to ignore. In December 2025, a federal court ordered OpenAI to produce 20 million ChatGPT conversations to opposing counsel in a copyright case. Three months later, the order expanded to cover another 88 million conversations. None belonged to parties in the litigation; they belonged to ordinary users who had typed into a system they experienced as private.

The conversation also explored a growing split among U.S. courts over whether using consumer AI tools can waive ordinary work-product protection, recent Supreme Court arguments over reverse-keyword warrants that could, if extended to chatbots, identify everyone who issued a prompt to an AI, and Meta’s launch of Incognito Chat with Meta AI on WhatsApp, which runs inference inside a hardware-isolated environment that Meta itself cannot inspect.

Participants considered whether conversations with AI assistants should receive protections analogous to those afforded communications with lawyers, doctors, therapists, or clergy. They examined the possible legal foundations for such protections, the limits they might require, and the technical safeguards necessary to make meaningful promises of confidentiality.

The discussion also addressed Sam Altman’s call for an “AI privilege.” While no statute or appellate decision has yet recognized such a privilege, speakers explored whether a credible framework would require more than a provider’s promise not to look at user conversations, and whether providers must be able to demonstrate that they could not compromise user confidentiality even under legal compulsion.

The webinar examined the four principal ways chatbot conversations can leave the confidential space users expect: voluntary corporate disclosure, compelled government access, civil discovery, and data breaches. It also explored emerging technical architectures—including confidential computing, client-side encryption, and “Sealed Mode”—that may reshape what providers can credibly promise and what the law can sensibly protect.

The event featured a moderated dialogue followed by audience Q&A and brought together lawyers, technologists, regulators, journalists, and interested members of the public.

Speakers:

Mikołaj Barczentewicz is a senior scholar at the International Center for Law & Economics and an associate professor of law at the University of Surrey. He writes EUTechReg and has argued that hardware-backed confidential computing—exemplified by Meta’s Incognito Chat—is the most credible path to combining frontier-grade AI with reasonable data safety. He also argues that voluntary technical self-restraint strengthens, rather than displaces, the case for legal protection. 

Theodore Christakis is a professor of international, European, and digital law at Université Grenoble Alpes. He is the author of “You Trust Your Chatbot With Everything. Should You?,” the first integrated treatment of the four pathways described above, including a detailed reconstruction of the NYT v. OpenAI discovery battle and the case for a “Sealed Mode” architecture.