Are AI Chats Privileged? What Heppner and Warner Mean for Lawyers and Clients

By The Professions Cloud Editors Updated October 2026 · 5 sources · 3 min read

Prices and facts checked October 13, 2026. We don't accept payment for rankings.

What happened: in February 2026, Judge Jed Rakoff of the Southern District of New York held in United States v. Heppner that 31 documents a criminal defendant created with a consumer AI tool, analyzing his own legal exposure, were protected by neither attorney-client privilege nor work product (BakerHostetler). The same month, a federal court in Michigan reached the opposite result for a pro se plaintiff's ChatGPT work (Cleary Gottlieb).

Why it matters today: clients increasingly ask AI tools about their legal problems before, or instead of, calling you. Those chats can be discoverable. How you advise clients about AI now affects what the other side may see later.

Heppner: the facts

• After receiving a grand jury subpoena and hiring counsel, the defendant used the consumer version of Claude to generate 31 documents on his exposure, defenses and strategy.
• He later shared them with his lawyers, who conceded he created them on his own, not at their direction.
• Bench ruling February 10, 2026; written opinion February 17.

Sources: Venable; BakerHostetler.

Why privilege failed

Sources: Venable; Orrick.

Read it narrowly or broadly? Venable stresses the court applied traditional, technology-neutral principles and didn't declare AI incompatible with privilege. Orrick reads it more broadly, warning that AI analysis of regulatory exposure may not be protected even on internal tools. That extension is untested. Notably, the court also suggested that if counsel had directed the AI use, the tool might be treated like a professional agent within the privilege (Dayon Torts blog).

Warner v. Gilbarco: the other answer

In the Eastern District of Michigan, Magistrate Judge Anthony Patti held that a pro se plaintiff's ChatGPT queries and responses were protected work product: they reflected her mental impressions in anticipation of litigation, and sharing with ChatGPT wasn't disclosure to an adversary (Cleary Gottlieb).

The difference: she was acting as her own counsel, so her litigation thinking was the work product. Heppner had lawyers and acted without them.

What to tell clients

Add this to your engagement letters and kickoff calls:

1. Don't use public AI tools to analyze your case, your exposure or our advice.
2. Don't paste our emails or memos into AI tools, since that can waive privilege.
3. If you want AI help, ask us first: we can direct it, using tools with proper confidentiality terms.
4. Assume anything you've already typed is discoverable, and tell us about it.

How firms should structure AI use

• Enterprise tools only for client matters: no training on inputs, defined retention, confidentiality terms (Venable).
• Document counsel's direction when clients or staff use AI on a matter.
• Update Upjohn warnings and investigation protocols to cover AI (Orrick).
• Treat AI outputs as potentially discoverable in preservation and collection.

Your own duties when using AI are covered in ABA Formal Opinion 512, and California lawyers face new statutory rules from 2027 (SB 574).

FAQ

Are enterprise AI tools privileged?

Unsettled. Strong confidentiality terms and counsel's direction help, but no appellate court has ruled.

Does this apply to civil cases?

Heppner was criminal, but the privilege analysis applies generally. Warner was civil.

Should I stop using AI for client work?

No. Use tools with proper confidentiality terms, direct their use and document it.

Changelog: Oct 13, 2026: first published.