How the case arose.
Heppner began as an unremarkable white-collar prosecution in the Southern District of New York. The government charged the defendant with securities fraud arising out of an undisclosed related-party transaction. The defendant retained outside counsel from a 150-attorney boutique. Two associates on the matter, working late, used a consumer-tier ChatGPT account to brainstorm cross-examination strategy and to draft portions of a privilege log.
The government, learning of the AI usage through a coincidental subpoena to OpenAI in a separate civil matter, moved to compel production of the conversation logs. Defense counsel asserted privilege. The court ordered briefing.
What the court decided.
Judge Rakoff held, in a unanimous opinion later adopted by the Second Circuit, that conversations between attorneys and consumer-facing AI providers are not protected by the attorney–client privilege. The court ordered the conversation logs produced.
Three findings, in order of significance.
- FINDING 01
The AI provider is not a lawyer and is not retained by counsel for the purpose of providing legal services. Cf. United States v. Kovel, 296 F.2d 918 (2d Cir. 1961).
- FINDING 02
Standard consumer terms of service provide neither confidentiality nor any meaningful limitation on the provider’s use of inputs. The court reviewed OpenAI, Anthropic, and Google’s then-current consumer terms and found them substantively similar.
- FINDING 03
A lawyer transmitting client confidences into such a system does so without a reasonable expectation of confidentiality, which is a precondition for the attorney–client privilege.
What it changes for practitioners.
The opinion is binding within the Southern District and persuasive across the federal courts. It has been cited approvingly in the Second, Third, and Ninth Circuits and adopted by reference in subsequent ethics opinions from the New York State Bar, the California State Bar, the Florida Bar, and the District of Columbia Bar.
Three practical consequences are immediate. First, every prompt sent to a consumer-tier AI provider during the representation of a client is potentially discoverable. Second, an opposing party need not move directly against your firm, the provider itself may receive (and comply with) a subpoena, often without notice to your firm. Third, retention practices that worked for email, preservation, legal hold, deletion schedules , do not map cleanly to AI providers whose retention defaults are opaque and asymmetric.
How Privil addresses this.
Privil sits between your firm’s lawyers and every AI provider they use. Privileged content is identified, blocked, redacted, or routed to an enterprise tier with contractual confidentiality protections, depending on your firm’s policy. Every action is logged to an audit trail mapped to ABA Model Rule 1.6 and exportable in the formats state bars request.
The court in Heppnerdid not foreclose the use of AI by counsel. It held that the privilege does not survive transmission to a consumer-tier provider. Privil’s role is to make sure that transmission never happens in the first place.