Tuesday night, 10 o’clock.
A Dallas taqueria owner just watched their taquero slip on a wet kitchen floor and be rushed away by ambulance. A workers’ compensation claim is imminent, with a potential lawsuit to follow.
The owner’s attorney specializes in commercial leases, not litigation. So, before making any calls, the owner does what many Texans now do instinctively: They open ChatGPT. The owner inputs details about what happened (who was in the kitchen, the conditions, the timeline), consults on questions for counsel and generates a summary, which they save on Google Docs and print in anticipation of a conversation with legal counsel the next morning.
Six months later, the owner is served with a petition and discovery request, demanding all documents created with any large language model related to the incident. During the deposition, the owner is asked whether they used an LLM after the accident and answers honestly. The document they created that night — intended as a proactive, AI-assisted account of the facts — becomes a key piece in opposing counsel’s case.
That result is not hypothetical anymore. In June 2026, a Texas Business Court confronted the work-product status of ChatGPT conversations in Tate Group Automotive LLC v. Legacy Automotive Capital LLC. After reviewing the chats in camera, Judge Grant Dorfman found that several logs were protected work product and rejected the argument that using ChatGPT itself waived that protection because the materials had not been disclosed to an adversary or in a way likely to reach one. The court also recognized the limits of that ruling, leaving for separate treatment the question of whether discovery materials had been uploaded to the tool in violation of a protective order. As a non-final, minute entry, the ruling signals how the Business Court approaches the question rather than settling it.
Tate is important, but the harder question is what comes next. Its doctrinal mechanics, including how Texas’s work-product rule diverges from the federal cases and what its arrival means for forum strategy, have already drawn able commentary in these pages. This article addresses a different, more practical question. Tate arose from sophisticated business litigation involving counsel, discovery productions and a protective order. Most Texans will encounter the issue in a less structured way: before counsel is retained, before a protective order exists and before they understand that a saved AI-assisted summary may become the subject of a discovery fight. The question is whether work-product protection will reach them.
That practical reality matters. The access-to-justice gap driving consumer use of AI is real. Estimates indicate that 92 percent of substantial civil legal needs among low-income Americans go unmet, and legal aid organizations can serve only about 30 percent of those who seek their help, according to a study by Colleen Chien and Miriam Kim. Chief Justice Roberts has acknowledged the stakes directly, noting that for those who cannot afford a lawyer, AI can help by providing accessible tools that answer basic legal questions without requiring them to leave home.
For Texas lawyers, the question now carries immediate practical urgency. Clients are already using AI in the same way as the taqueria owner, and discovery requests are beginning to target those uses. Whether that protection will reach the individuals and small businesses most likely to use consumer AI before counsel is retained is the question this article addresses.
A trend is emerging where courts are rejecting the automatic-waiver theory: the view that typing facts into a consumer AI tool automatically waives work-product protection. AI use clearly does not eliminate discovery obligations: Underlying facts remain discoverable, platform terms may matter and protective orders must be obeyed. But, as the reasoning goes, using a consumer AI tool to prepare for litigation by itself does not forfeit whatever work-product protection Texas law would otherwise provide. But there is another concern that has largely been unaddressed. If Texas treats ChatGPT differently from Google Docs, email, cloud storage, e-discovery platforms, and other third-party legal technology, work-product protection may turn less on litigation purpose than on purchasing power.
Interface, Not Infrastructure
The automatic-waiver argument depends on treating a new interface as a new kind of disclosure. It runs like this: When a party types facts into ChatGPT, the party has sent information to a third-party server; because the information reached a third party, work-product protection is lost. That argument has surface appeal only if AI is treated as technologically exceptional. But is it really?
Return to the taqueria owner. The owner did not only use ChatGPT. They also saved the summary in Google Docs and printed it for counsel. If the document had been drafted entirely in Google Docs, without the use of AI, no one would say work-product protection vanished because the text passed through Google’s servers. If the owner emailed the same summary to counsel, no one would say that the email provider had become a litigation adversary. The work-product question would turn on why the document was created, who created it and whether it was later disclosed in a way inconsistent with protection. Is ChatGPT any different?
Modern litigation preparation already depends on third-party infrastructure. Litigation holds, document review and legal research often run through cloud-based vendors. Yet lawyers do not ordinarily treat those tools as disclosure to a litigation adversary.
The question turns on whether AI, as new software, is treated any differently from old software. The privilege question has focused on whether disclosure to a vendor is tantamount to disclosure to an adversary or through a channel substantially likely to place the material in the adversary’s hands. Platform choice can raise confidentiality, retention, cybersecurity, contractual and protective-order issues. Those issues are real. But the use of a software tool has never worked as an automatic work-product waiver.
Tate illustrates that separation. The court protected some ChatGPT logs while preserving the distinct question of whether discovery materials had been uploaded in violation of a protective order. Where a party uses a nonpublic AI interface to organize facts, prepare questions or frame issues for counsel, the mere involvement of a software provider, treated as the legal equivalent of handing the document to an opponent, establishes a curious new paradigm.
A Two-Tier Rule
Under this new paradigm, an automatic-waiver rule would not affect all litigants equally. Sophisticated companies can procure enterprise AI platforms, negotiate confidentiality commitments, require data isolation and route AI use through counsel-supervised workflows. The individuals and small businesses most likely to rely on consumer tools cannot do any of that.
That divide is not abstract. The Texans most likely to open a consumer chatbot before seeing a lawyer are the same people least able to retain one. The access-to-justice figures describe them: the small-business owner with no litigation counsel and the individual facing a claim who cannot afford an hourly rate. An automatic-waiver rule would seem to fall on the very people who turned to AI because earlier legal help was out of reach, converting a tool of access into a trap and only for those who could not buy their way around it.
The practical consequences are partially predictable. Lawyers may tell clients not to use consumer AI at all, even to organize facts before a consultation. Opposing counsel would be motivated to serve broad requests for every prompt, output, chat log, and AI-assisted draft related to a dispute. Courts would be, as they already have been, asked to decide waiver before they ever reached the ordinary work-product questions of litigation purpose, substantial need, undue hardship, and adversary disclosure.
These consequences would not yield a neutral rule. Such a rule would protect the party with the budget to use enterprise tools and expose the party using the consumer tools most available to ordinary Texans. It would also discourage the conduct lawyers should want clients to undertake before the first meeting: writing down timelines, preserving details, identifying witnesses, organizing documents and preparing focused questions for counsel.
Under the traditional landscape, the owner who writes a timeline by hand, types it into Google Docs or asks ChatGPT to help organize it for counsel would face the same analytical framework regardless of which tool they used. The relevant questions would be whether the material was prepared in anticipation of litigation, whether the underlying facts can be obtained through ordinary discovery and whether the material was disclosed in a manner inconsistent with protection.
The new paradigm may turn work-product protection into a function of purchasing power. If AI tools are becoming part of how clients prepare for legal advice, the reasoning must consider the distinction between careless disclosure and ordinary preparation. An automatic waiver collapses that distinction, appearing to protect the work product of parties who can pay for confidentiality and stripping it from everyone else. Work product protection should not belong to only those who can afford to keep it.
The Limits Are Practical, Not Categorical
Texas does not need a new category of privilege to reach that result. Courts should consider these practical implications as well as the scope of the rules as they have traditionally existed. Rule 192.5 already protects material prepared or mental impressions developed “in anticipation of litigation or for trial by or for a party or a party’s representatives.” The Tate court protected the ChatGPT logs under that text, treating them as material a party prepared in anticipation of litigation. The point is straightforward: The rule asks why the material was created, not whether the first draft was written on a legal pad, in Microsoft Word, in Google Docs or through an AI interface.
The rule also contains explicit limiting principles. A party invoking work-product protection must show, under the totality of the circumstances, that a reasonable person would have anticipated litigation and that the party actually prepared the material for that purpose, the standard the Texas Supreme Court set in National Tank Co. v. Brotherton. Ordinary business records, routine incident reports, customer-service notes and general compliance materials do not become work product merely because someone later asks an AI tool to summarize them. Underlying facts remain discoverable, non-core work product may be discoverable on a showing of substantial need and undue hardship and core work product remains reserved for an attorney’s or representative’s mental impressions, opinions, conclusions or legal theories. This is also work-product protection, not attorney-client privilege: A consumer chatbot is not a lawyer, and no attorney-client privilege attaches to a conversation with one. Work product is different because it can survive some third-party exposure when the disclosure is not to an adversary and is not substantially likely to reach one.
The federal decision pointing the other way, Heppner, turned on different reasoning and facts, denying protection because the materials were not prepared by or for counsel and treating the doctrine as a shelter for the lawyer’s thinking rather than the party’s. Texas Rule 192.5 answers that reasoning directly. It protects material prepared by or for a party, not only the lawyer, and the Tate court read it that way.
The practical response is to ask better questions early. At the outset of a matter, counsel should ask whether the client used any AI tool to organize facts, draft timelines, summarize documents, prepare questions for counsel or analyze the dispute. If the answer is yes, counsel should identify what was entered, what was generated, when the exchange occurred, what platform was used and whether any confidential discovery material or protective-order material was included.
Counsel should also document the litigation purpose where it exists. If a client created an AI-assisted chronology, summary or question list because a claim or lawsuit was anticipated, that fact should be preserved in the file. Doing so does not manufacture work-product protection after the fact, but it will help courts apply Rule 192.5 to the actual circumstances in which the material was created.
Lawyers should also give clients clear instructions on the use of AI. Clients should not upload attorney-client communications, produced discovery, confidential business information, trade secrets, medical records, personally identifying information or attorney’s-eyes-only material into consumer AI tools without counsel’s approval. Protective orders should expressly address AI tools, including whether confidential materials may be processed through third-party platforms, under what conditions and with what safeguards.
That guidance preserves the right to at least bring those distinctions before the court and argue their paramount relevance to the privilege question. AI use may raise confidentiality, ethics, cybersecurity, retention and protective-order issues, which should be managed directly. Tate reinforces that principle and rejects a categorical waiver rule. The next step is making that principle workable for the clients most likely to use these tools before counsel is in place.
Rogelio Reyes is a commercial litigator at Platt Richmond PLLC in Dallas who litigates complex disputes in Texas state and federal courts, with a practice that includes sports, arts, entertainment, contract, compliance, and document-intensive business disputes. He built his practice at an AmLaw 20 firm and on secondment with a professional NFL franchise, experience that sharpened his command of high-stakes disputes at the intersection of business, media and regulated industries. A graduate of the University of Texas School of Law and a native Spanish speaker, he serves on the board of directors of Dallas Contemporary.
