Does Using AI Waive Attorney-Client Privilege? Breaking Down the United States v. Heppner Case

The short answer: no. But the case does have some important lessons for how you guide your clients.

When United States v. Heppner made headlines, reactions ranged from concerned to alarmed. Does consulting an AI chatbot destroy privilege? Is legal AI no longer safe to use? Do lawyers need to stop using these tools entirely?

The ruling is more nuanced than the reactions suggest. Given that it was issued by Judge Jed Rakoff, one of the most influential federal judges in the country, it’s worth reading closely. The decision does have implications for your practice, just not the ones many commentators described.

The facts behind United States v. Heppner

The defendant, Heppner, was under federal indictment for wire fraud, securities fraud, and conspiracy to commit securities fraud. When investigators executed a search warrant on his home, they seized 31 documents among other materials.

His counsel flagged those documents as protected, either under attorney-client privilege or as work product, and asked the court to exclude them from evidence.

The problem? The documents were printouts from the free version of Claude. Heppner had apparently gone to the publicly available, free-tier AI chatbot, which can train on user data, typed in his facts, asked for defense strategies, and printed out the results. No law firm account. No paid subscription. And critically, no involvement from his attorney.

Judge Jed Rakoff ruled that neither privilege nor work-product protection applied.

The attorney-client privilege analysis

For attorney-client privilege to apply, the facts must include three elements: (1) a communication between a client and their attorney, (2) that was intended to be—and in fact was—kept confidential, and (3) made for the purpose of obtaining or providing legal advice. Rakoff found that Heppner’s Claude printouts failed on at least the first two, and likely all three.

• No attorney involved. Claude is not a lawyer. A conversation with an AI chatbot is not a communication with an attorney. This was the clearest basis for the ruling.
• No confidentiality. The free version of Claude may train on user data. There is little or no confidentiality promise, no data protection agreement, and no reasonable expectation that the information shared would remain private. 
• Purpose of obtaining legal advice. Rakoff was less definitive here, but Claude’s terms of service include a disclaimer that it is not a lawyer and cannot provide legal advice. That cut against Heppner’s argument that he was seeking legal counsel.

The work product analysis

Work product doctrine protects materials prepared in anticipation of litigation that reflect the mental processes of an attorney. On this point, Rakoff focused on a key requirement: The materials must be prepared by or at the direction of counsel.

Heppner’s own attorney admitted at the hearing that the lawyer had never asked Heppner to generate these documents, or to search Claude in the first place. The printouts had no connection to defense counsel’s strategy or mental processes, and without that connection, work product protection couldn’t apply.

What the ruling is not

This is where some of the online commentary went wrong.

Rakoff did not hold that using AI waives privilege, that AI-assisted legal work is unprotected, or that lawyers who use generative AI tools lose work product protection.

What he held is more narrow: a represented client, acting alone and without his attorney’s knowledge, used a free public AI tool with no confidentiality protections, and the resulting printouts are not privileged.

A lawyer using a secure, purpose-built legal AI tool, one with zero data retention, proper confidentiality protections, and SOC 2 Type 2 certification, would face none of the problems in this case. The confidentiality element is satisfied. The connection to the attorney’s mental processes is present. Work product protection applies, just as it would with any other legal research tool.

How U.S. v. Heppner impacts your firm

The Heppner case is less about AI and more about a client who went off-script, and a lawyer who didn’t have the right guardrails in place to prevent it. Here are the practical takeaways.

• Talk to your clients about AI. Many clients may not know that using a free AI chatbot to brainstorm their legal strategy is the rough equivalent of discussing their case with a barber. The communication isn’t privileged. Now that Heppner is on the books, those conversations can be discoverable. Make sure your clients understand this before it becomes a problem.
• Direct clients to the right tools if they want to do their own research. If it makes sense for a client to do preparatory research between meetings, tell them what to use, and put your instructions in writing so you can show a judge later if necessary. A paid, privacy-protected AI option with a proper confidentiality policy is a very different product from a free public chatbot, and that distinction now has a court opinion behind it.
• When done right, your own AI use is protected. When you use a secure legal AI tool as part of your practice—to research, draft, and analyze—that work is privileged. The Heppner ruling doesn’t change that. Instead, it clarifies that the protection flows from the attorney relationship, the confidentiality controls, and the connection to legal strategy. 
• Document how AI fits into your workflow. As courts continue to interpret privilege in the context of AI, having a clear, consistent practice around which tools you use and why will matter. A written AI policy for your firm, even a simple one, demonstrates intentionality and helps protect both you and your clients.

Where AI and privilege law goes from here

Heppner is one of many cases that will likely shape privilege and work product in our AI era. Courts will draw distinctions based on the specific tools used, whether an attorney was involved, what confidentiality protections were in place, and how the materials were generated. 

There are also harder questions the Heppner ruling doesn’t answer. What about self-represented litigants who use AI for legal research? Pro se plaintiffs, individuals who represent themselves in court without an attorney, and jailhouse lawyers have historically been able to claim some work product protection. If a defendant who can’t afford a lawyer turns to AI for help preparing their case, should those materials be discoverable in a way that a represented defendant’s notes wouldn’t be? That’s a real access-to-justice concern, and one that future courts will need to address.

In fact, just a few weeks before Heppner, a federal magistrate judge in Michigan held that a pro se party’s use of AI was protected as work product. Warner v. Gilbarco, Inc., No. 2:24‑cv‑12333-GAD-APP, Docket #94 (E.D. Mich. Feb. 10, 2026). Just as the past few years have seen myriad “hallucinated cases” sanctions, we’ll likely see “AI work product” cases for years to come, too.

For now, the clearest lesson from Heppner is a simple one: the privilege protections lawyers and clients have always relied on remain intact when AI is used properly, within the attorney-client relationship, using legal AI with appropriate confidentiality controls.

Lawyers who use purpose-built legal AI tools like Clio Work are protected. They’re better equipped to research thoroughly, identify issues early, and deliver stronger outcomes for their clients. For lawyers using legal AI the right way, this ruling changes nothing, except perhaps the conversation you need to have with your clients.