United States v. Heppner (S.D.N.Y. 2026): AI Chatbot Documents Not Privileged
Issued by
United States District Court for the Southern District of New York (Judge Jed Rakoff)
- September 30, 2026 · Correction — Rewrote the entry: a February 2026 criminal-case ruling, Heppner's actual roles, and the decisive lack of counsel direction (fact-check finding). (Cody Maxwell)
- October 1, 2026 · Correction — Rewrote the practical steps, newsletter hook, search description, and audience fields to match the entry's corrected content. (Cody Maxwell)
In United States v. Heppner, Judge Jed Rakoff of the Southern District of New York ruled in February 2026 that 31 documents a criminal defendant created with Anthropic's Claude were not privileged. Heppner, former CEO of Beneficient and chairman of GWG Holdings, had used consumer Claude on his own, not at his lawyers' direction.
Applies To
Overview
Bradley Heppner, founder and former CEO of Beneficient and chairman of GWG Holdings, faces federal fraud charges. After his arrest in November 2025, FBI agents searched his home and seized devices holding about 31 documents from his exchanges with Anthropic's Claude. The government moved for a ruling that they were not privileged. Judge Jed Rakoff ruled from the bench on 10 February 2026 and issued a written opinion on 17 February 2026. Because Claude is not a lawyer, there was no attorney-client relationship, and the court rejected treating Claude as a mere tool. Heppner used the consumer version, whose privacy policy lets Anthropic collect, train on, and disclose inputs and outputs. So he had no reasonable expectation of confidentiality. Counsel's role was decisive here. Heppner conceded he had not used Claude at his lawyers' direction, which also defeated work product protection. The judge suggested the result might differ if counsel had directed the AI use. Commentators have since asked how the reasoning fits with routine use of email and cloud documents, which courts have not treated as waiving privilege. The ruling binds only this case but is persuasive elsewhere.
Key Requirements
- •A client's own use of a consumer AI chatbot for legal questions was not protected by attorney-client privilege.
- •The documents were not work product, because they were not prepared by or at the direction of counsel.
- •The consumer service's privacy policy, which allows data collection and disclosure, defeated any expectation of confidentiality.
- •The court suggested AI use directed by counsel might be treated differently.
- •Organizations should set rules on when staff may use AI tools for legal matters, and prefer tools with confidentiality terms.
What Your Organization Must Do
- →Tell staff that personal use of consumer AI chatbots for legal questions may not stay confidential.
- →Route AI-assisted legal work through counsel, since the court suggested lawyer-directed use may be treated differently.
- →Review the privacy terms of AI tools, and prefer services that limit collection, training and disclosure.
- →Write an internal rule on which AI tools may touch legal matters, and which are off limits.
- →Brief litigation and e-discovery teams that chatbot transcripts may be collectible evidence.
- →Note that this decision binds only this case, but other courts may find it persuasive.
Playbook Guidance
Step-by-step implementation guidance for compliance teams.
Frequently Asked Questions
- Does the Rakoff ruling mean all AI-generated legal documents are discoverable in federal litigation?
- No, the ruling covered a client's own use of a consumer chatbot without counsel's direction. Use directed by counsel, or tools with strong confidentiality terms, could be analyzed differently.
- Which AI platforms were specifically at issue in the Rakoff ruling?
- The case involved the consumer version of Anthropic's Claude. Its privacy policy, which permits collecting, training on, and disclosing data, was central to the finding of no expected confidentiality.
- Does this ruling bind courts outside the Southern District of New York?
- No, it binds only this case. It is persuasive authority, and law firms are already citing it in advice on privilege and AI-assisted legal work.
- What were the consequences of the ruling?
- This was a criminal case, and the government already held the documents from a search. The ruling let it use them; in civil cases the risk is that similar documents are not shielded from discovery.
- Does it matter whether counsel directed the use of the AI chatbot?
- It matters a great deal. Heppner used Claude on his own, and the judge suggested the result might differ if counsel had directed the AI use.
- How does this ruling compare to existing case law on privilege and third-party email servers?
- The court did not rely on an email analogy. Commentators raise it as an open question, since courts have not treated lawyers' use of email or cloud documents as defeating privilege.