United States v. Heppner

In United States v. Heppner (February 2026), Judge Jed Rakoff of the Southern District of New York ruled that conversations with consumer AI tools aren't protected by attorney-client privilege. The defendant used Claude — the consumer version, not an enterprise deployment — to research legal strategies without his lawyer's knowledge or direction. The prosecution moved to compel production of those AI chat logs, and the court agreed.

This isn't a close call. Judge Rakoff gave three independent reasons the privilege doesn't apply, any one of which would've been enough. The ruling is a blueprint for prosecutors seeking AI chat logs and a warning for every defense attorney who hasn't updated their engagement letters. If your client is talking to an AI about their case, those conversations are discoverable.

Judge Rakoff's Three Reasons Privilege Doesn't Apply

Judge Rakoff's ruling was systematic. First, Claude isn't an attorney — privilege requires communication between a client and a licensed lawyer. An AI chatbot, regardless of how sophisticated its legal analysis appears, doesn't qualify. The court rejected the argument that AI functioning "like" a lawyer satisfies the privilege requirement.

Second, Anthropic's Terms of Service explicitly state that user conversations may be used for model improvement and safety monitoring. By agreeing to those terms, the defendant waived any expectation of confidentiality. You can't claim privilege over a communication you've already authorized a third party to access.

Third, the defendant used Claude on his own initiative, not at the direction of counsel. Privilege only extends to agents and intermediaries when they're facilitating attorney-client communication at a lawyer's direction. Self-directed AI research doesn't qualify.

The Kovel Doctrine Exception Left Open

The most important part of Heppner is what Judge Rakoff didn't rule on. The court explicitly left open whether the Kovel doctrine — which extends privilege to communications with non-lawyer experts working at an attorney's direction — could cover AI tools used under counsel's supervision.

Under Kovel (established in *United States v. Kovel*, 296 F.2d 918, 2d Cir. 1961), communications with accountants, translators, and other experts are privileged when those experts are engaged by the attorney to assist in rendering legal advice. Judge Rakoff noted that a "properly structured AI consultation directed by counsel" might satisfy Kovel's requirements — but the facts in Heppner didn't come close.

This leaves a narrow path for privilege protection: the attorney must select the AI tool, direct the specific queries, use an enterprise deployment with no third-party data access, and document the consultation as part of the legal representation. Anything less and you're in Heppner territory.

Why Every Firm Needs Updated Engagement Letters Now

Heppner creates an immediate obligation for defense attorneys: tell your clients that talking to consumer AI about their case destroys privilege. This isn't optional guidance — it's a malpractice risk. If a client uses ChatGPT or Claude to research their legal situation and opposing counsel subpoenas those logs, everything the client told the AI is fair game.

Engagement letters need three additions after Heppner. First, an explicit prohibition on discussing case facts with consumer AI tools. Second, a list of firm-approved AI tools (if any) that have been vetted for privilege protection. Third, a clear statement that unauthorized AI consultations may waive privilege over the information disclosed.

Firms using AI internally need separate protocols. Enterprise AI deployments with no-training agreements, access controls, and attorney-directed workflows may survive a Kovel analysis. Consumer tools with default Terms of Service won't.

Implications for Criminal Defense and Investigations

The prosecution in Heppner obtained the AI chat logs through a subpoena to Anthropic, which complied under its standard legal process policy. This means prosecutors now have a proven playbook: subpoena the AI provider, get the chat logs, use them against the defendant.

For criminal defense attorneys, the exposure is massive. Defendants regularly research their charges online — that's expected. But AI conversations are qualitatively different from Google searches. Clients ask Claude specific questions about their situations, disclose facts, and explore defense strategies. Those conversations read like privileged attorney-client communications, but after Heppner, they're not.

The ruling also applies to civil litigation. Any party using consumer AI to analyze their legal position is creating discoverable evidence. Plaintiffs' attorneys should be adding AI chat logs to their standard discovery requests immediately.

How to Structure AI Use to Preserve Privilege

Heppner doesn't ban AI from legal practice — it defines the boundaries. To preserve privilege, firms need a documented framework. The attorney must direct the AI interaction, choosing specific prompts and controlling the scope. The AI tool must be deployed under an enterprise agreement that prohibits data retention and third-party access.

The interaction must be documented as part of the attorney's work product — not a separate, client-initiated conversation. And the firm must maintain written policies that were communicated to the client before the AI was used. Judge Rakoff's opinion all but spelled out these requirements in his Kovel analysis.

Firms that get this right gain a significant advantage — AI-assisted legal analysis that's both faster and privileged. Firms that don't are creating discoverable evidence every time they open a chat window.

The Bottom Line: After Heppner, any client conversation with consumer AI is discoverable — update your engagement letters, lock down your AI protocols, and tell clients to stop talking to chatbots about their cases.