Sept. 2, 2026 — The promise that artificial intelligence (AI) could help self-represented parties creates a danger that lawyers interviewing new clients need to ask about.

This year, a small number of cases across the country, starting in federal courts, weighed whether self-represented litigants may use AI and keep their searches confidential as work product.

These cases warn that AI searches by represented clients may not be protected.

Lawyers need to know whether prospective clients asked AI about their legal issue — and to make sure clients don’t prompt about their case into AI.

Two Federal Cases

The first word came Feb. 10 in Warner v. Gilbarco, Inc., a discovery dispute seeking “all documents and information concerning [plaintiff’s] use of third-party AI tools in connection with this lawsuit.”^[1]

The court denied the request. Because the plaintiff was self-represented, use of AI constituted work product.

Entering the information into ChatGPT didn’t waive privilege, the court explained, because the information wasn’t likely to reach the adversary.

“And ChatGPT (and other generative AI programs) are tools, not persons, even if they may have administrators in the background.”

United States v. Heppner, signed a week after Warner, went the opposite way, but so did the facts.^[2]

Heppner involved a criminal defendant represented by counsel.

Bradley Heppner used AI after he received a grand jury subpoena. He knew he was the target of an investigation.

His lawyer didn’t know that Heppner accessed AI, including inputting information he learned from his lawyer. Heppner later shared the AI results with his lawyer.

The AI documents failed to fit within the work product doctrine because “they were nevertheless not ‘prepared by or at the behest of counsel,’ … nor did they reflect defense counsel’s strategy,” the court explained.

The court held that “communicat[ing] with a publicly available AI platform in connection with a pending criminal investigation” is not protected work product.

Morgan may be more useful than Heppner. The difference is deeper than whether one has a lawyer. It’s between realities of civil litigation and criminal prosecution.

“Notably, the work product doctrine applies more narrowly in the criminal context than in civil litigation,” lawyers have explained.^[3]

Criminal defendants also have a constitutional right to representation. They will more likely have a lawyer.^[4]

Self-representation Equalizer?

“Both parties appear to be using AI in connection with their litigation work,” the court recognized in Morgan v. V2X, Inc. The business sought information about plaintiff’s AI to protect confidential business information.^[5]

The court held the information was off limits as work product.

Changes to Rule 26 in 1970, the Advisory Committee explained, extended work product “protection beyond attorneys’ work product to materials prepared by or for a party,” including a pro se litigant, the court emphasized.

While pro se litigants benefit from these protections, policy reasons encouraged the court.

“The importance of applying these protections to pro se litigants is magnified in the context of AI — one of the most powerful knowledge tools ever to become available to the masses. This is because pro se litigants are forced to act as both party and advocate, simultaneously.”

“And for the first time in history, widespread access to powerful technology may make that dual role surmountable.”

As “[p]ro se litigants are held to the same standard as represented litigants,” the court explained, “[t]hey should be afforded the same protections.”

Although AI systems collect user data, “nearly all electronic interaction passes through third-party systems” in which one may have a reasonable expectation of privacy, the court described.

“[I]t is highly unlikely the information will fall into the hands of an adversary absent some legal process to compel it.”

Morgan advised the Nassau County, New York trial court to protect a pro se defendant’s “prompts, inputs, uploaded materials, and corresponding outputs or generated texts” related to the case.^[6]

The court found persuasive Morgan’s analysis of Warner’s assertion that AI submissions did not waive confidentiality.

This line of cases offers a new route to access justice.

“[F]or those who cannot afford a lawyer, AI can help,” suggested U.S. Supreme Court Chief Justice John Roberts.

“It drives new, highly accessible tools that provide answers to basic questions, including where to find templates and court forms, how to fill them out, and where to bring them for presentation to the judge — all without leaving home.”^[7]

Such advice may be available from free legal clinics, operating in various locations and with limited schedules — if the individual knows about them.

Unrepresented and self-represented parties do not receive work product protections in all jurisdictions.^[8] These few cases released do not extend that protection.

‘Romantic Partner’ Work Product?

The key to protection remains in self-representation, although the bigger problem in Shealy v. Seaside Investments, LLC in Suffolk County Superior Court’s Business Litigation Session involved more than having a lawyer.^[9]

In a dispute about an agreement for purchase and sale of a gas station, Earl Wade Shealy, Jr., asked “his romantic partner” for legal advice from ChatGPT.

In response, Seaside Investments sought “the queries of AI programs for legal assistance by a party’s romantic partner” and “the AI output.”

No case or rule “has held that a party’s romantic partner is his or her ‘representative’ such that documents prepared by that individual are protected by the work product doctrine,” the Massachusetts trial court explained.

“Plaintiffs are represented by sophisticated, capable counsel,” which didn’t ask Shealy to proceed with these communications. These facts were “more akin to” the facts in Heppner. The information was discoverable.

‘Explicit Warnings to Clients’

Although the rapid development of AI proclaims something new, some of these problems are old wine in new bottles.

Echoing Judge Frank H. Easterbrook at the dawn of the consumer internet, recent commentary advises “we must be careful to avoid … ‘AI exceptionalism’ — treating AI as so different and special that we fail to see how the privacy problems with AI are the same as existing privacy problems, just enhanced.”^[10]

Just as American Bar Association (ABA) Model Rule 1.1 note 8 expanded to specify a continuing duty of competence “including the benefits and risks associated with relevant technology,” AI demands the same duties.

“[T]he duty of competence should now include anticipating how a client’s personal use of AI … may expose privileged communications or strategy,” wrote District of Columbia Superior Court Senior Judge Herbert B. Dixon Jr. last spring.^[11]

Competence may also require lawyers to always seek discovery on AI tools, prompts, and responses, whether out of zealous advocacy or to protect proprietary client information.^[12]

Dixon advocated for the ABA Standing Committee on Ethics and Professional Responsibility to revisit ABA Formal Opinion 512 — its first word in 2024 on the ethical duties of lawyers using AI.

“The opinion should expressly require lawyers to provide proactive guidance and explicit warnings to clients about the risks inherent in using AI tools independently while their case is pending,” Dixon advised.^[13]

That advice should happen at intake, when providing copies of pleadings, and when litigation holds begin — including educating employees about discoverability of AI-generated content.^[14]

Highlighting the duty to explain to clients the legal effects of AI will protect clients from themselves — to seek out more information using readily available computer tools and thereby make discoverable what they don’t want the other side to know.

Ethics Watch is a monthly article that tracks ethics trends and developments nationwide. It is NOT reviewed or written by the State Bar of Wisconsin’s ethics counsel attorneys, who write Ethical Dilemmas.

Endnotes

[1] Warner v. Gilbarco, Inc., 820 F.Supp.3d 629, 636-37 (E.D. Mich. 2026); Herbert B. Dixon Jr., When Is a Litigant’s AI Research Discoverable?, 65 No. 2 Judges’ J. 37 (Spring 2026) (“At the time the Heppner and Warner decisions were issued, there had been no appellate resolution of similar AI discovery-related issues in any U.S. state or federal appellate court.”). ↩

[2] United States v. Heppner, 820 F.Supp.3d 292, 294, 297-99 (S.D. N.Y. 2026). ↩

[3] John P. Carlin, et al., Federal Courts Reach Different Outcomes on Whether AI-Generated Materials Warrant Work Product Protection, Insights: Client Memorandum, Paul, Weiss, Rifkind, Wharton & Garrison LLP, New York and Washington, D.C., Mar. 25, 2026, https://www.paulweiss.com/insights/client-memos/federal-courts-reach-different-outcomes-on-whether-ai-generated-materials-warrant-work-product-protection (last visited Aug. 20, 2026). ↩

[4] U.S. Const. amend. VI; Wis. Const. art. I, § 7; Anoo Dinesh Vyas, Confidentiality of AI Conversations: Protecting Self-Represented Litigants Who Use ChatGPT for Legal Advice, 26 Duke L. & Tech. Rev. 186, 202 (Apr. 17, 2026) (“Work product disputes for self-represented litigants may not arise as often in the criminal context because defendants may have a constitutional right to an attorney when facing imprisonment.”). ↩

[5] Morgan v. V2X, Inc., No. 25-cv-01991-SKC-MDB, *2-*5, *8 (D. Colo. Mar. 30, 2026), available at 2026 WL 864223. ↩

[6] Assini v. Hayward, ___ N.Y.S.3d ___ (Nassau County N.Y. Sup. Ct. 2026), available at 2026 WL 1677232 and https://www.nycourts.gov/reporter/current/3dseries/2026/2026_26086.shtml (last visited Aug. 20, 2026). ↩

[7] Vyas, supra note 4, at 193. ↩

[8] Id. at 201. ↩

[9] Shealy v. Seaside Invs., LLC, No. 2684CV00799-BLS2 (Suffolk County Mass. Super. Ct. June 16, 2026), available athttps://masslawyersweekly.com/files/2026/07/09-070-26.pdf (last visited Aug. 20, 2026); Matthew W. Costello & Jack Tierney, Lessons from Shealy v. Seaside: When Client AI Use Can Sink Work Product Protection, Now & Next: Litigation & Technology Alert, Nixon Peabody, LLP, Boston, July 29, 2026, https://www.nixonpeabody.com/-/media/files/alerts/2026/07/lessons_from_shealy_v_seaside_when_client_ai_use_can_sink_work_product_protection.pdf (last visited Aug. 20, 2026) (providing additional details). ↩

[10] Vyas, supra note 4, at 221 (referencing Frank H. Easterbrook, Cyberspace and the Law of the Horse, 1996 U. Chi. Legal F. 207 (1996)). ↩

[11] Dixon, supra note 1, at 39. ↩

[12] Jennifer Colegate, AI — The Quiet Erosion of Privilege?, Insights: Briefings, Baker & Partners LLP, Cayman Islands, July 15, 2026, https://www.bakerandpartners.com/insights/ai-the-quiet-erosion-of-privilege/ (last visited Aug. 20, 2026) (explaining that “consumer-grade AI models retain user inputs” for training software upgrades, and “a client’s proprietary trade secrets or litigation strategy could theoretically resurface in a response generated for a competitor”). ↩

[13] Dixon, supra note 1, at 39. ↩

[14] Costello & Tierney, supra note 9, at 4 (providing “practical takeaways”). ↩