Blab AI
Federal Courts Clarify How Generative AI Affects Legal Privilege and Discovery
When a defendant casually taps into a chatbot, the courtroom can become a battlefield for confidentiality.
In March 2026, a judge in the Southern District of New York answered the question: do conversations with a generative‑AI tool fall under attorney‑client privilege or the work‑product doctrine? The ruling, United States v. Heppner, 820 F. Supp. 3d 292 (S.D.N.Y. 2026), held that documents a defendant drafted with Anthropic’s Claude chatbot were not protected. Bradley Heppner had used the public platform on his own to outline defense strategies for a federal criminal investigation, accepting the terms of service that allow the AI to use user inputs for model training or third‑party review. Judge Jed Rakoff found that because Heppner acted alone and voluntarily disclosed the information to a third‑party AI provider, there was no reasonable expectation of confidentiality, and the materials were not privileged.
The case sparked discussion because it implied that a consumer‑grade AI tool could strip away otherwise protectable communications. Yet its scope was narrow: a single criminal defendant and a specific public platform.
A month later, a different court offered a broader perspective. In Morgan v. V2X, Inc., No. 25‑CV‑01991‑SKC‑MDB, 2026 WL 864223 (D. Colo. Mar. 30, 2026), Magistrate Judge Maritza Dominguez Braswell addressed the same privilege questions in a civil‑litigation context. The parties were engaged in discovery, had protective orders, and the AI tool was used under counsel’s supervision. The court shifted focus from the mere presence of AI to the safeguards surrounding its use. It noted that lawyers routinely employ third‑party services—email, cloud storage, e‑discovery platforms—without losing privilege, provided those services are engaged to facilitate representation and not to share information with adversaries. Emphasizing contractual protections, the ability to delete data, and compliance with protective orders, the decision concluded that privilege is preserved when the AI is used responsibly and within established confidentiality safeguards.
A third decision turned the spotlight to discoverability. In Conservation Law Foundation v. Shell Oil Co., No. 3:21‑CV‑00933 (VDO), 2026 WL 764396 (D. Conn. May 18, 2026), the court held that the prompts and queries an expert used to prepare a report were discoverable under Rule 26(b) of the Federal Rules of Civil Procedure. Because the expert’s methodology relied on those prompts, the inputs were considered part of the expert’s analysis.
Taken together, the three cases paint a coherent picture. Heppner shows that careless use of a public AI platform can waive privilege. Morgan demonstrates that privilege can survive when the AI is used under proper safeguards. Conservation Law reminds practitioners that the content of AI prompts can become part of the discoverable record.
For law firms, the practical implications are clear:
1. Review the platform’s terms of service to ensure user inputs are not used for model training. 2. Prefer enterprise agreements that contractually restrict data retention and third‑party access. 3. Supervise AI use and document the safeguards in place. 4. Align AI workflows with any protective orders or ethical obligations. 5. Preserve and document expert prompts, as they may be subject to discovery.
These decisions signal a broader trend: courts are treating AI as another tool that must be governed by existing confidentiality and discovery rules, rather than as a new threat to those rules. As more cases arise, the framework established by Heppner, Morgan, and Conservation Law is likely to guide how privilege, work product, and discovery are applied to AI‑generated content.
The decisions are still early in the legal landscape, and appellate courts have not yet weighed in. Nevertheless, they provide the first substantive guidance on how generative AI interacts with long‑standing legal doctrines, offering a roadmap for attorneys navigating the growing use of AI in practice.
In March 2026, a judge in the Southern District of New York answered the question: do conversations with a generative‑AI tool fall under attorney‑client privilege or the work‑product doctrine? The ruling, United States v. Heppner, 820 F. Supp. 3d 292 (S.D.N.Y. 2026), held that documents a defendant drafted with Anthropic’s Claude chatbot were not protected. Bradley Heppner had used the public platform on his own to outline defense strategies for a federal criminal investigation, accepting the terms of service that allow the AI to use user inputs for model training or third‑party review. Judge Jed Rakoff found that because Heppner acted alone and voluntarily disclosed the information to a third‑party AI provider, there was no reasonable expectation of confidentiality, and the materials were not privileged.
The case sparked discussion because it implied that a consumer‑grade AI tool could strip away otherwise protectable communications. Yet its scope was narrow: a single criminal defendant and a specific public platform.
A month later, a different court offered a broader perspective. In Morgan v. V2X, Inc., No. 25‑CV‑01991‑SKC‑MDB, 2026 WL 864223 (D. Colo. Mar. 30, 2026), Magistrate Judge Maritza Dominguez Braswell addressed the same privilege questions in a civil‑litigation context. The parties were engaged in discovery, had protective orders, and the AI tool was used under counsel’s supervision. The court shifted focus from the mere presence of AI to the safeguards surrounding its use. It noted that lawyers routinely employ third‑party services—email, cloud storage, e‑discovery platforms—without losing privilege, provided those services are engaged to facilitate representation and not to share information with adversaries. Emphasizing contractual protections, the ability to delete data, and compliance with protective orders, the decision concluded that privilege is preserved when the AI is used responsibly and within established confidentiality safeguards.
A third decision turned the spotlight to discoverability. In Conservation Law Foundation v. Shell Oil Co., No. 3:21‑CV‑00933 (VDO), 2026 WL 764396 (D. Conn. May 18, 2026), the court held that the prompts and queries an expert used to prepare a report were discoverable under Rule 26(b) of the Federal Rules of Civil Procedure. Because the expert’s methodology relied on those prompts, the inputs were considered part of the expert’s analysis.
Taken together, the three cases paint a coherent picture. Heppner shows that careless use of a public AI platform can waive privilege. Morgan demonstrates that privilege can survive when the AI is used under proper safeguards. Conservation Law reminds practitioners that the content of AI prompts can become part of the discoverable record.
For law firms, the practical implications are clear:
1. Review the platform’s terms of service to ensure user inputs are not used for model training. 2. Prefer enterprise agreements that contractually restrict data retention and third‑party access. 3. Supervise AI use and document the safeguards in place. 4. Align AI workflows with any protective orders or ethical obligations. 5. Preserve and document expert prompts, as they may be subject to discovery.
These decisions signal a broader trend: courts are treating AI as another tool that must be governed by existing confidentiality and discovery rules, rather than as a new threat to those rules. As more cases arise, the framework established by Heppner, Morgan, and Conservation Law is likely to guide how privilege, work product, and discovery are applied to AI‑generated content.
The decisions are still early in the legal landscape, and appellate courts have not yet weighed in. Nevertheless, they provide the first substantive guidance on how generative AI interacts with long‑standing legal doctrines, offering a roadmap for attorneys navigating the growing use of AI in practice.
