United States v Heppner

United States v. Heppner (Case No. 25-cr-00503-JSR, S.D.N.Y.) is a criminal fraud case in which Judge Jed S. Rakoff ruled that a defendant's documents generated using the publicly available Anthropic Claude AI tool are not protected by attorney-client privilege or the work product doctrine. Issued on the same day as Warner v Gilbarco — which reached the opposite conclusion — the paired rulings represent the first judicial determinations on whether AI-generated materials qualify for legal privilege protections.

Parties

Plaintiff

• United States of America — prosecuting federal fraud charges

Defendant

• Bradley Heppner — criminal defendant who independently used Claude AI to generate litigation documents before his arrest

Court

• Court: United States District Court for the Southern District of New York
• Judge: Jed S. Rakoff
• Case No.: 25-cr-00503-JSR

Factual Background

After receiving a grand jury subpoena and learning he was a target of a federal fraud investigation, Heppner independently used the non-enterprise, publicly available Claude AI platform — without counsel's direction — to generate 31 documents. He input information received from his lawyers into Claude and later shared the outputs with counsel for legal advice. Federal agents seized electronic devices containing these materials during a search of Heppner's mansion.

Ruling

Oral Ruling (February 10, 2026)

Judge Rakoff orally granted the government's motion to compel from the bench on February 10, 2026, stating there was "not remotely any basis for any claim of attorney-client privilege."

Written Opinion (February 17, 2026)

The court issued a written opinion on February 17, 2026, ordering production of the 31 documents.

Legal Analysis

Attorney-Client Privilege — Rejected

The court denied privilege on three grounds:

1. No communication between client and attorney — Heppner communicated directly with Claude AI, not with his counsel. The AI is not a lawyer and the interaction is not a privileged communication.
2. No reasonable expectation of confidentiality — Claude's terms of service disclaim legal advice and confidentiality, meaning the AI platform functions as a third party. Inputting information into a public AI tool effectively discloses it to a third party.
3. Not created for obtaining legal advice from counsel — Heppner used Claude independently, without counsel's direction. Sharing outputs with counsel afterward does not retroactively create privilege.

The court noted that inputting counsel-provided information into Claude may constitute a waiver of privilege over those original attorney-client communications.

Work Product Doctrine — Rejected

The court rejected work product protection because the documents were created by Heppner on his own volition, without counsel's involvement or direction. Unlike a party acting through an attorney, an unrepresented criminal defendant's independent AI use does not qualify as work product prepared "in anticipation of litigation" under counsel's supervision.

Kovel Framework

The court referenced the Kovel framework, leaving open the possibility that privilege could apply if counsel had directed the client to use AI before the interaction — similar to how an accountant's work under an attorney's direction can be privileged. The ruling does not foreclose privilege for attorney-supervised AI use.

Comparison with Warner v. Gilbarco

On the same day (February 10, 2026), a federal magistrate judge in Warner v Gilbarco reached the opposite conclusion, protecting AI-assisted materials under the work product doctrine. Key distinctions:

Factor	Heppner (S.D.N.Y.)	Warner (E.D. Mich.)
Case type	Criminal	Civil employment
Counsel status	Had counsel, but acted independently	Pro se (acting as own counsel)
Privilege	Denied — no attorney-client communication	Not extended (but not needed)
Work product	Denied — no counsel direction	Protected — pro se status allows assertion
AI as tool vs. person	Treated AI interactions as third-party disclosures	"AI tools are tools, not persons"

Significance

United States v. Heppner is the first reported ruling addressing whether AI-generated materials are protected by legal privilege. Key implications:

• No automatic privilege for AI use — Using a public AI tool without counsel involvement does not create privilege or work product protection
• Waiver risk — Inputting privileged information into a public AI tool may waive privilege over that information
• Safe harbor for directed use — Counsel-directed AI use may qualify for privilege under the Kovel framework
• Enterprise vs. public AI — The ruling specifically addressed publicly available Claude; enterprise AI tools with confidentiality assurances may receive different treatment
• Criminal vs. civil context — The criminal setting heightened the court's skepticism toward self-generated privilege claims

Case Information

• Court: U.S. District Court for the Southern District of New York
• Case No.: 25-cr-00503-JSR
• Judge: Jed S. Rakoff
• Filed: 2025
• Ruling Date: February 10, 2026 (oral); February 17, 2026 (written opinion)

See Also

• Warner v Gilbarco — Paired ruling protecting AI-assisted work product
• News Article on United States v. Heppner
• Cases — List of AI-related litigation