William Slomanson
Distinguished Professor Emeritus
Thomas Jefferson School of Law
Email: bills@tjsl.edu
AI and privilege waiver
Court decisions involving generative artificial
intelligence focus on misuse of AI by attorneys (and some judges). For case
examples, sanctions, and legislative quests, see Slomanson, "AI
Hallucinations and the Bar," 39 Cal. Litigation Rptr. 1 (forthcoming, 2026).
The following AI case illustrates how a client can waive any privilege. The Feb.
10, 2026, Memorandum decision by a New York federal district court is the
nation's first case to address the novel question of whether a party's
interaction with AI waives the attorney-client and work product privileges. U.S.
v. Heppner, ____ F.Supp.3d ____, 2026 WL 436479 (S.D. NY, Feb. 17, 2026)
[hereinafter Heppner].
Before dissecting this novel decision, the novelty of AI
merits a brief comment about the associated technology. An AI system is
one "that can, for a given set of human-defined objectives, make predictions,
recommendations, or decisions influencing real or virtual environments." "Artificial
Intelligence & Responsible Business Conduct," OECD (2019), p. 1,.
Within that system, "'generative artificial intelligence' means ... [a
model] that can generate derived synthetic content, including text, images,
video, and audio that emulates the structure and characteristics of the
system's training data." Cal. Senator Tom Umberg, Legislative Council's Digest,
"Generative artificial intelligence: attorneys and arbitrators," Senate Bill
574 § 2(B) (Jan. 30, 2026), [hereafter
SB-574].
Heppner is an executive in several corporate entities. He pled not guilty to
federal fraud charges. The FBI executed a search warrant at Heppner's home and
seized numerous documents and electronic devices. Thirty-one of those
documents memorialize communications that Heppner had with Anthropic's virtual
assistant Claude. Without any suggestion from counsel that he do so, Heppner
thereby prepared reports that outlined defense strategy, e.g., what he
might argue with respect to the facts and the law in anticipation of a
potential indictment. Heppner, *1. The Government and Heppner's counsel
agreed that the prosecutors would segregate the AI documents, but not inspect
them, pending resolution of Heppner's privilege claims. Id., *2.
The Heppner trial judge granted the Government's request, seeking to
access those documents. They were not protected from government inspection via
either the attorney-client or work product limitations on discovery. Heppner,
*1. Heppner is not alone. He is one of the planet's estimated 800 million
people who initiate Anthropic AI searches each week. Over half of this nation's
residents have access to Claude. See Mary Meerker, et al., Trends_Artificial_Intelligence.pdf"Trends_Artificial_Intelligence,"
at 59 & 55.
The core of the attorney-client privilege claw back
argument was that "(1) Heppner had inputted into Claude ... information that
Heppner had learned from counsel; (2) Heppner had created the AI Documents for
the purpose of speaking with counsel to obtain legal advice; and (3) Heppner
had subsequently shared the contents of the AI Documents with counsel." Heppner,
*2. The court reasoned that "the AI Documents lack ... [the usual] elements of
the attorney-client privilege.
First, the AI Documents are not communications between Heppner and his
counsel. ... [T]he discussion of legal issues between two non-attorneys is not
protected by attorney-client privilege." Id. The judge nevertheless
considered a related assertion: "some commentators have argued that whether
Claude is an attorney is irrelevant because a user's AI inputs, rather than
being communications, are more akin to the use of other Internet-based software,
such as cloud-based word processing applications." Id. The court's read,
however, was that "all '[r]ecognized privileges' require, among other things,
'a trusting human relationship,' such as, in the attorney-client context, a
relationship 'with a licensed professional who owes fiduciary duties and is
subject to discipline.'" Id., citing Distinguished Professor of Law Ira
Robbins, "Against
an AI Privilege," JOLT Digest, Harvard Law School (Nov. 7,
2025).
Second, the communications memorialized in the AI documents were not
confidential. Heppner had communicated with third-party Claude. Furthermore,
Anthropic's written privacy policy is that Anthropic collects data on user
inputs and Claude's outputs, over which Anthropic reserves the right to
disclose such data to third parties--specifically including governmental
regulatory authorities (citing Anthropic's privacy policy). Heppner, *2.
Third, Heppner did not communicate with Claude for the purpose of obtaining
legal advice. But the court acknowledged that "[t]his issue perhaps presents a
closer call because Heppner's counsel asserts that Heppner communicated with
Claude for the 'express purpose of talking to counsel.'" Id., *3. So,
the privilege door is not sealed shut, where counsel directs a client to
consult an AI platform. Thus, "[h]ad counsel directed Heppner to use Claude,
Claude might arguably be said to have functioned in a manner akin to a highly
trained professional who may act as a lawyer's agent within the protection of
the attorney-client privilege."
The final coffin nail was that "AI Documents 'would not be
privileged if they remained in [Heppner's] hands,' [and] they did not 'acquire
protection merely because they were transferred' to counsel." Id., 3.
Likewise in California: "that which was not privileged in the first instance
may not be made so merely by subsequent delivery to the attorney." Chadbourne,
Inc. v. Superior Court of the City and County of San Francisco, 60 Cal.2d
723, 732 (Cal., 1964).
The use of electronic communications does not modify
precedent. A "communication between persons in a relationship ... does not lose
its privileged character for the sole reason that it is communicated by
electronic means or because persons involved in the delivery,
facilitation, or storage of electronic communication may have access to the
content of the communication." Cal. Evid. Code § 917(b). But California's
Evidence Code provisions are never broadly applied--as
evinced by the Heppner result. This litigant's unwitting attempt to
expand the attorney-client privilege regime was a fool's errand. The legal
terrain lowers that shield "[b]ecause privileges 'prevent the admission of
relevant and otherwise admissible evidence,' [so] they should be narrowly
construed." People v. Sinohui, 28 Cal.4th 205, 212 (Cal., 2002). Heppner's
ploy, notwithstanding today's expanding technological applications to the law,
presented a bridge too far.
The Heppner work product claim suffered a like
rejection. The court reasoned that "The[se] AI Documents do not merit
protection under the work product doctrine because, even assuming, arguendo,
that they were prepared 'in anticipation of litigation,' they were
nevertheless not 'prepared by or at the behest of counsel,' nor did they
reflect defense counsel's strategy." Heppner, *4. Nor was Heppner acting
as his lawyer's agent when he was communicating with Claude. Counsel stated
that this defendant prepared the referenced documents via his own, independent
misadventure.
Counsel's third strike hypothesized that the AI documents
his client generated with Claude may be privileged because they "incorporated
information that we had conveyed to Heppner over the course of our
representation." Id., *4 n.3. But even if that information was in
fact privileged, Heppner waived the privilege by sharing it with Anthropic.
Claude was akin to a third party who was not in the immediate attorney-client
loop.
The legal horizon is now fraught with AI hallucinations; a
new perspective about attorney sanctions; and now, Heppner's privilege
waiver via generative AI. Counsel should advise clients that, when they
communicate with virtual AI assistants, their "work product" is fair game for
discovery. One hopes that unlike Heppner, we lawyers don't fall prey to
Proverbs 23:8 --"A poor workman blames his tools."