AI Chats and Discovery: A Warning for Texas Family Lawyers
By Patrick A. Wright1
1 Patrick A. Wright is a graduate of South Texas College of Law. He is Board Certified in Family Law and Child Welfare by the Texas Board of Legal Specialization and is a member of the American Academy of Matrimonial Lawyers. He is a Co-Editor of this Section Report, a Co-Chair for the ABA TECHSHOW 2026, was a Co-Course Director for the 2025 Advanced Family Law Course, a member of the LPM Committee of the State Bar of Texas, was recently selected to be a council member for the ABA Law Practice Division Council, and the Managing Partner of The Wright Firm, L.L.P. in Flower Mound, Texas. Patrick’s involve-ment with various committees, speaking, and writing credits focuses on AI, future initiatives by both the Family Law Section and the ABA, and the practice of family law. He may be reached at Patrick@thewrightlawyers.com.
A February 2026 article in The National Law Review reports a significant fed-eral discovery and privilege ruling out of the Southern District of New York: on Feb-ruary 10, 2026, Judge Jed Rakoff ordered production of a criminal defendant’s AI chat materials in United States v. Heppner.2 No. 25 CR. 503 (JSR), 2026 WL 436479, at *1–2 (S.D.N.Y. Feb. 17, 2026). The defendant had used a non-enter-prise, consumer version of Anthropic’s Claude to research legal issues, generated documents consisting of prompts and responses, and later sent those materials to his attorneys—after which the FBI seized them during a search. When the govern-ment moved to compel, the defense asserted attorney-client privilege and work-product protection. Judge Rakoff rejected both.
2 Andrew R. Lee, Jason M. Loring & Graham H. Ryan, Your AI Conversations Are Not Privileged: What a New SDNY Ruling Means for Every Lawyer and Client, Nat’l L. Rev. (Feb. 13, 2026), https://natlawreview.com/article/your-ai-conversations-are-not-privileged-what-new-sdny-ruling-means-every-lawyer
Even though Heppner arises in federal criminal litigation, the reasoning is immediately relevant to Texas family lawyers because it frames AI prompts and outputs as ordinary third-party communications and ordinary ESI, not privileged legal consultations. For practitioners who are already seeing clients use AI to draft timelines, “summaries,” parenting narratives, custody proposals, and financial explanations, the decision is a warning: a client’s “private” chat with an AI platform may be the opposite of private in discovery.
The federal logic (FRCP, FRE, and why privilege failed)
According to the article, Judge Rakoff accepted the government’s arguments that privilege failed for four core reasons. First, no attorney was involved—an AI tool is not a lawyer and cannot form an attorney-client relationship or owe professional duties. Second, the communications were not for the purpose of obtaining legal advice, and the platform’s materials disclaimed providing legal services—undercutting any claim that the user sought legal advice from the tool itself. Third—and most practically important—the communications were not confidential. The article highlights that the platform terms and policies con-templated disclosure to governmental or regulatory authorities and use for training, which the court treated as eliminating a reasonable expectation of confidentiality. Fourth, pre-existing documents cannot be retroactively cloaked: sending unprivileged AI materials to counsel after the fact did not convert them into privileged communications.
Work product did not fare better. The article reports that defense counsel conceded the defendant created the AI documents “of his own volition” and that counsel did not direct him to run the AI searches. Without attorney direction (and without materials created “by or for” counsel), the work-product claim failed. The article also notes an “unexpected wrinkle”: because the AI materials incorporated information counsel conveyed to the client, using them at trial could create witness-advocate complications if counsel had to testify about what they told the client.
Finally, the article flags a waiver risk: feeding attorney communications into an AI tool can be treated as sharing privileged material with a third party, potentially waiving privilege over the original attorney-client communications.
Translating the lesson to Texas family law (TRE and TRCP)
Under Texas Evidence and Procedure, privilege is built on confidential communications made to facilitate legal services, and work product still depends on litigation context and lawyer-directed prepara-tion. If a client independently generates AI prompts/outputs on a consumer platform, those materials look
much less like protected attorney work product and much more like discoverable “documents” and “data compilations” created and stored with a third party.
In a Texas divorce or SAPCR, this matters because AI use often overlaps with the most litigated issues: credibility, conservatorship, possession, safety narratives, substance use allegations, and the money story. A client’s AI chat log could become impeachment gold: inconsistent versions of events, strategic framing (“what should I say in court”), or ill-advised questions (“how do I hide assets”), even when the client’s intent was simply anxiety-driven “research.” Heppner underscores the risk that courts will treat these logs as discoverable ESI—because, from a discovery standpoint, that is what they are.
What Texas family lawyers should do now
The article’s practical recommendations translate cleanly into Texas family-law risk management:
1. Client onboarding script + written warning. Tell clients, explicitly, not to input facts, allegations, strategy, finances, or child-related details into consumer AI tools—because it may be discoverable and may not be privileged.
2. Ask the AI question early. Add to your initial interview checklist: “Have you used ChatGPT/Claude/Gemini/etc. about this case? Any saved chats or exports?”
3. Control the firm’s AI use. The article distinguishes standard consumer accounts from enterprise agreements with contractual confidentiality protections; treat AI like any other vendor risk area, and document guardrails accordingly.
4. Preservation and waiver discipline. If a client already used AI, assume the logs exist some-where. Manage preservation thoughtfully and avoid compounding waiver by further third-party disclosures.
By contrast, in Warner v. Gilbarco, Inc., the court declined to compel production of AI-generated drafting materials created during litigation. No. 2:24-CV-12333, 2026 WL 373043, at *4 (E.D. Mich. Feb. 10, 2026). The court held that the plaintiff’s use of ChatGPT for internal analysis and drafting did not waive work-product protection. Id. Critically, the court characterized generative AI as a “tool[],” not a “person[],” and concluded that waiver of work product requires disclosure to an adversary or conduct substantially increasing the likelihood that the material would reach an adversary. Id. (emphasis omitted). Because the AI-generated materials remained internal and were not shared with opposing counsel, pro-tection was preserved.
Read together, Heppner and Warner do not announce conflicting rules; rather, they delineate the contours of existing doctrine in the AI context. Independent AI use by a party, untethered to counsel’s direction and lacking reflection of attorney strategy, is unlikely to receive work-product protection and will not qualify for attorney–client privilege. By contrast, AI employed as part of an internal, attorney-directed litigation workflow may fall within traditional work-product safeguards—so long as the material is not dis-closed in a manner inconsistent with maintaining its confidentiality.
Bottom line: Heppner is a reminder that AI chat feels private while often functioning like a third-party disclosure. For Texas family lawyers, that gap—between client perception and discovery reality—can determine whether a “research session” becomes the other side’s best exhibit.