A Texas business court has allowed a litigant to withhold most of his ChatGPT conversations as attorney work product, finding that the chats contained material and mental impressions developed by a party in anticipation of litigation. The decision offers an early state-court view of how traditional discovery protections may apply when a person uses a generative AI system to analyze an active legal dispute.
In Tate Group Automotive v. Legacy Automotive Capital, Kris Tate used ChatGPT while working through issues connected to litigation, according to an analysis published by Texas Lawbook. Tate Group Automotive withheld the resulting conversations during discovery, asserting work product protection. The defendants sought to compel their production, arguing both that a nonlawyer’s interaction with an AI chatbot could not qualify as work product and that submitting information to ChatGPT would waive any protection.
Judge Grant Dorfman of the Business Court of Texas reviewed the conversations privately, in camera, rather than deciding their status solely from the parties’ general descriptions. He concluded that most of the material could remain withheld, while directing the production of particular portions that did not qualify for protection. According to Texas Lawbook, the court found that the protected conversations fell within Texas Rule of Civil Procedure 192.5(a)(1), which covers material prepared or mental impressions developed in anticipation of litigation or for trial by or for a party or the party’s representatives.
The ruling is significant because it treats the litigant’s own litigation preparation as potentially protected even when an attorney did not direct the work and the analysis was conducted through a consumer AI service. The court’s reasoning focused on the purpose and substance of the material, rather than treating the use of a chatbot as automatically disqualifying.
That approach contrasts with the federal decision in United States v. Heppner. As summarized by Texas Lawbook, the federal court in that matter declined to protect AI-generated documents that a financial executive prepared on his own before consulting counsel. Although the materials were assumed to have been created in anticipation of litigation, the court determined that they had not been prepared by counsel, at counsel’s direction or as a reflection of counsel’s legal strategy.
The two decisions illustrate that disputes over AI-assisted litigation materials involve at least two separate legal questions. A court must first decide whether the material is work product at all. If it is, the court must then determine whether the user surrendered that protection by providing information to an outside AI platform. Different doctrinal answers can lead to different outcomes even when litigants engage in broadly similar conduct.
On the first question, the Texas court read the state rule to encompass work performed by a party, not only work undertaken by an attorney. According to Texas Lawbook, Dorfman found that most of Tate’s conversations appeared on their face to contain litigation-related material or mental impressions prepared by a party. The fact that Tate used ChatGPT instead of recording his thoughts through a more conventional medium did not, by itself, remove the material from the rule.
The court also rejected the argument that using ChatGPT necessarily waived work product protection. In doing so, it followed reasoning identified in the federal cases Warner v. Gilbarco and Morgan v. V2X. Those decisions distinguish disclosure to an adversary, or conduct that substantially increases the likelihood of adversarial access, from the use of an outside technology provider. Under that approach, entering information into a chatbot does not automatically amount to the type of disclosure that forfeits work product protection.
That conclusion does not establish that consumer AI tools are confidential in every legal sense. AI services can retain prompts, process them through external infrastructure or reserve contractual rights concerning data use and disclosure. The Texas ruling instead addresses the narrower doctrine of waiver, which generally focuses on whether protected material was exposed to an opponent or placed in circumstances that make acquisition by an opponent substantially more likely.
The decision also does not transform a conversation with ChatGPT into an attorney-client communication. Attorney-client privilege protects qualifying confidential communications made for the purpose of obtaining or providing legal advice between a client and a lawyer. A chatbot is not a lawyer, and a person independently asking it legal questions does not thereby form an attorney-client relationship. Work product is a separate protection centered on preparation for litigation, and its waiver rules can be more tolerant of some third-party involvement than the rules governing attorney-client privilege.
Even after prevailing on most of the work product issue, Tate did not receive complete protection from discovery concerning his AI use. Texas Lawbook reported that the court required disclosure of the AI tool and the discovery materials supplied to it, including information that had been produced under a protective order. The substance of most of Tate’s strategic exchanges could remain withheld, but the fact of using the tool and the inputs drawn from discovery were not treated as categorically protected.
That distinction creates an important technical and procedural issue for litigants. A party using generative AI on discovery material may need to identify exactly what was uploaded, pasted or otherwise submitted. If that activity is not logged or documented, reconstructing it later could be difficult. The use of material designated confidential, attorneys’ eyes only or as a trade secret may also raise obligations independent of work product doctrine.
Dorfman reserved judgment on whether submitting protected discovery to the AI platform violated the governing protective order, according to Texas Lawbook. The court encouraged the parties to discuss amendments addressing the use of AI systems and large language models. Such provisions can define which services are permitted, whether submitted data may be retained or used for model training, how highly sensitive documents must be handled, and what notice or recordkeeping is required.
The ruling’s procedural status limits its authority. Texas Lawbook described it as a nonfinal minute entry issued through the Business Court’s local discovery process, rather than a final appellate ruling. A party seeking a definitive decision may still pursue formal motion practice, including a motion to quash, for protection or for reconsideration. The result therefore indicates how this court analyzed the materials before it, but it does not settle the issue across Texas or establish binding appellate precedent.
Still, the dispute shows why the forum and the wording applied by a particular court can matter when AI-generated or AI-assisted records surface in discovery. Courts are beginning to adapt established doctrines to tools that can combine a user’s confidential factual inputs, litigation theories and generated responses in a single transcript. Those records may resemble personal notes, research memoranda, communications with a service provider or some mixture of all three, depending on how the system was used.
The Texas decision supports a document-specific inquiry rather than a blanket rule that all chatbot conversations are protected or discoverable. It also leaves practical limits in place: litigants may have to reveal which system they used, account for what they gave it and comply with protective orders governing third-party access. Until appellate courts provide broader guidance, parties handling sensitive litigation information through generative AI remain exposed to differing outcomes based on jurisdiction, procedural posture and the contents of each conversation.
Sources: texaslawbook.net