THE HEPPNER PROBLEM

1The Heppner Problem describes the class of professional exposure created by United States v. Heppner, No. 1:25-cr-00503-JSR (S.D.N.Y. Feb. 17, 2026): consumer AI platform interactions carry no attorney-client privilege, no work product protection, and no legally cognizable confidentiality. Documents generated through those interactions are discoverable. The data underlying those interactions has been ordered produced by courts in parallel litigation.

2The term applies beyond the criminal defendant facts of the original case. Any professional who has used a consumer AI platform for work that touched confidential matters (legal, tax, healthcare, financial, regulatory) faces a version of the same exposure.

3A legal doctrine that protects confidential communications between a client and a licensed attorney made for the purpose of obtaining or providing legal advice. The privilege belongs to the client and can be waived by the client.

4Three elements are required: (1) a communication between a client and an attorney, (2) that is intended to be and was kept confidential, (3) for the purpose of obtaining or providing legal advice. Failure on any one element defeats the privilege. In Heppner, Judge Rakoff held the 31 AI-generated documents failed at least two of the three elements.

5A legal doctrine that protects materials prepared by counsel, or at counsel's direction, in anticipation of litigation or for trial. The core purpose is to protect the mental processes and strategy of attorneys. Unlike attorney-client privilege, work product protection is not limited to communications. It extends to documents prepared by counsel or at counsel's direction.

6The key requirement is attorney direction. Materials prepared by a client independently, without attorney direction, do not qualify. Heppner's counsel conceded that his attorneys did not direct him to use Claude. That concession was fatal to the work product claim.

7An extension of attorney-client privilege to communications with nonlawyer agents whose assistance is necessary to facilitate the attorney-client relationship, derived from United States v. Kovel, 296 F.2d 918 (2d Cir. 1961). Under Kovel, privilege can extend to accountants, translators, technical experts, and others engaged by counsel to assist in the representation, provided the agent operates under counsel's direction and confidentiality is maintained.

8Judge Rakoff noted in Heppner that had counsel directed Heppner to use Claude, Claude might arguably have functioned as a Kovel agent. That question was not decided. Whether an AI platform used at counsel's direction, inside a system with contractual confidentiality terms, could qualify as a Kovel agent is an open question courts have not yet addressed.

9A prerequisite for attorney-client privilege. If a communication was made in circumstances where confidentiality could not reasonably be expected (because it was shared with a third party, made in a public setting, or transmitted through a channel with disclosed privacy limitations), privilege does not attach.

10Judge Rakoff held in Heppner that Anthropic's privacy policy, which expressly permitted collection and disclosure of user data, destroyed any reasonable expectation of confidentiality. The finding is not specific to Anthropic. OpenAI's privacy policy contains comparable provisions.

11A court order requiring a party to retain data that would otherwise be deleted or destroyed. Preservation orders are issued in litigation to ensure that potentially relevant evidence is not lost before discovery. They override a party's standard data retention and deletion policies, including commitments made to users.

12On May 13, 2025, U.S. Magistrate Judge Ona T. Wang issued a preservation order requiring OpenAI to retain all ChatGPT output log data that would otherwise be deleted, including user-deleted conversations and data subject to GDPR deletion requirements. The order was affirmed by District Judge Sidney Stein on June 26, 2025.

13A category of discoverable evidence under the Federal Rules of Civil Procedure. ESI includes emails, documents, databases, voicemail, and any other information stored in electronic form. Federal Rule of Civil Procedure 34 permits parties to request production of ESI from opposing parties. Courts can also order production of ESI held by third parties through subpoena.

14In the OpenAI copyright litigation, courts have held that ChatGPT conversation logs are discoverable ESI. In re OpenAI, Inc. Copyright Infringement Litigation (MDL, S.D.N.Y.). AI conversation logs are not treated differently from other forms of ESI for discovery purposes.

15A contractual arrangement between an enterprise customer and an AI platform provider under which the provider does not retain user inputs or outputs after the session ends. OpenAI offers ZDR agreements for enterprise API customers. Under a ZDR agreement, the platform holds no data to produce in response to legal process.

16OpenAI's May 2025 preservation order specifically excluded ChatGPT Enterprise and API customers with ZDR agreements. Consumer tiers (Free, Plus, Pro, Team) were covered by the order. Bloomberg Law noted that several products marketed for "business" or "enterprise" use offer no more legal protection than consumer services under Heppner's reasoning, and that the distinction between consumer and enterprise tiers requires review of the specific governing terms.

17The concept, applied by Judge Stein in the OpenAI copyright litigation, that ChatGPT users voluntarily submitted their communications to OpenAI when they used the platform. That voluntary submission distinguished their privacy interests from subjects of surreptitious wiretaps, who were recorded without knowledge or consent. The distinction was central to the court's rejection of OpenAI's privacy objection to producing the 20 million conversation logs.

18The voluntary submission analysis maps onto the confidentiality analysis in Heppner. In both cases, the user's choice to use the platform under its disclosed terms of service was held to limit their privacy expectations in the resulting data.

19For purposes of the Heppner analysis, a consumer AI platform is any publicly accessible AI tool whose terms of service permit the platform to collect user inputs and outputs for model training and to disclose that data to third parties in response to legal process. This includes ChatGPT Free, Plus, Pro, and Team tiers; Claude Free, Pro, and Max tiers; and comparable consumer-tier products from other providers.

20The designation is defined by the terms of service, not by price point. A $20-per-month subscription does not change the underlying terms. Bloomberg Law's analysis noted that some products marketed for business or enterprise use may have terms that offer no more protection than consumer services.