Does a Mediator Blow Privilege By Using AI?

The Question? Last week a California colleague forwarded me the following question. If I as the mediator put confidential case information into an AI tool, and it produces a timeline, would that timeline be considered protected by California’s mediation confidentiality laws?

In my view, in a non-criminal matter in a California court the answer is clear. Not so clear in most other states, or in federal court.

What Brought This Up? The question was triggered by the February 2026 decision in US v Heppner. The decision analyzes and addresses how the court should view a defendant who’s preparing for his case by uploading case documents into an AI, and then sending the AI’s product to his attorney.

<https://www.courtlistener.com/docket/71872024/27/united-states-v-heppner/>

I often get confidentiality-related questions, having played a central role in crafting the mediation confidentiality laws in California and a dozen other states. For several decades now, I’ve also been privileged to discuss with visionaries and pioneers the new technologies they were developing for the dispute resolution field.

Those comfortable with using AI suggest we treat this confidentiality question as we would a mediator using cloud-based word processing software or email. AI is simply another tool.

Skeptics suggest we view it more like a mediator sharing confidential information with persons outside the statutory protections for mediation communications.

I've been astonished and impressed by AI’s ability to craft original analysis and persuasive argument. I've also been alarmed at how effectively these can sometimes disguise serious factual errors, including AI-fabricated falsehoods. In my view, we’re better off working to understand AI as something new and rapidly evolving - neither like a tool nor like a human.

Court’s Decision in Heppner? Referring to the AI by name, the Court said “Because Claude is not an attorney…that alone disposes of Heppner's claim of privilege.” Some discussing this decision have cited this as the central point of the analysis.

But the Court’s analysis also relied on the widely-recognized theory of waiver by conduct. If I want to preserve the privileged status of communications with my attorney, I better not share those communications with other people. I may later be deemed by a court to have waived my attorney/client privilege.

If we treat disclosure to an AI like disclosing it to other people, a court may find this was waiver by conduct.

My Response to the Question? If it was created for the purpose of a mediation, the legislative intent in California was that it would be protected - whether that timeline was created by a human expert, a party, a mediator, or a program.

Is California Different? Yes. Even in California, experienced mediators, attorneys, and judges often refer to our statutory protections for mediation communications as privileges. But as the California Supreme Court reaffirmed in the Simmons case, unlike most other states California's protections for mediation communications are not structured as privileges.

Under California's statutory scheme, mediation communications are instead excluded from admissibility and disclosure on public policy grounds. Unlike attorney/client and similar privileges our protections are not waived if mediation communications are shared with third parties.

Simmons v. Ghaderi. As our Supreme Court explained in Simmons,

“Unlike the privileges subject to implied waiver that are found in division 8, entitled ‘Privileges,’ the Legislature placed section 1115 et seq. in division 9, entitled ‘Evidence Affected or Excluded by Extrinsic Policies.’ This placement reflects that the Legislature considered the specific limitations placed on the admissibility of evidence by the mediation confidentiality statutes and endorsed those limitations to encourage mediation as a matter of public policy.”

“Recognizing both the breadth and clarity of the mediation confidentiality statutes, we have concluded that the legislative scheme is clear and unambiguous, and that the Legislature intended for mediation confidentiality to apply according to the statutory rules.”

- Simmons v. Ghaderi (2008) 44 Cal.4th 570

 What’s Required in California? California’s basic protections for mediation communications are provided by our Evidence Code sections 1119 and 1121. Relevant portions of section 1122 then provide the general conditions under which these communications may be admitted or disclosed as follows -

“A communication…made or prepared for the purpose of…a mediation…is not made inadmissible, or protected from disclosure…if…All persons who conduct or otherwise participate in the mediation expressly agree in writing.”

In the Simmons case, even when both parties introduced mediation communications in open court, our Supreme Court determined they had not impliedly waived California's protections by their conduct. In this author's opinion, this was consistent with the legislative intent.

So When Is a Communication Protected? A key question of fact for a later reviewing California court will normally be whether a communication was “made or prepared for the purpose of…a mediation” and what evidence can properly be considered by the judge deciding this. Mediators, counsel, and parties may want to take care that their communications carry some clear indication they were made for the purpose of a mediation.

Criminal Proceeding? Federal Court? Two other points are worth remembering in California, since the Heppner case was a criminal proceeding in federal court. First, unlike some other states, California’s Evidence Code chapter governing mediation provides no protections for mediation communications in later criminal proceedings. Second, federal judges sometimes decline to treat mediation communications as protected even in cases where the Erie Doctrine would theoretically require admissibility to be governed by California or other state-based law providing clear protections.

Privilege-Based Statutes? What about people in states which use the Uniform Mediation Act’s or similar privilege structure to protect mediation communications? The Uniform Mediation Act also aims to protect communications “made for purposes of…a mediation.” A later factual determination on this question will likely always be central.

But as in the US v Heppner case, privilege-based protections have regularly been deemed waived by conduct.

Waiver by Conduct? Suppose you’re a mediator using AI in a privilege-based state. Which AI are you using? Is it a typical publicly-available AI? Did you read the terms of service? Did these say you consented to the company reviewing your communications with the AI, using them to further train the AI, and/or disclosing them to third parties, as is typical?

In a privilege-based state, that might be enough to convince a judge you blew the privilege.

If you used an AI system which assured you of confidentiality, that might mean a different ruling on the waiver question, but that's far from guaranteed.

As the judge in Heppner so aptly put it “Generative artificial intelligence presents a new frontier in the ongoing dialogue between technology and the law.”

Proceed with caution. The road ahead may be slippery.

** Since the mid-1980s, Ron Kelly has been actively involved in the formation of mediation law and policy. He initiated and guided enactment of California’s main Evidence Code chapter defining and governing mediation, and played a central role nationally in shaping the Uniform Mediation Act.

Ron’s trained thousands of judges, attorneys, and business professionals in mediation skills, law, and ethics on four continents. He’s been honored with eleven major awards for his work in building the field. More information on crafting sound mediation law and policy is here: https://www.ronkelly.com/pg6.cfm

Copyright Ron Kelly, 2026. Permission is granted to copy, post, and distribute this article unedited in its entirety - including the bio, copyright notice, and permission to distribute.