AI Discovery

Defendants sought discovery including whether prompts posed to generative AI (e.g., ChatGPT, Claude) are protected by the work product doctrine. The court, adopting the reasoning of Morgan v. V2X, held that AI interactions do not automatically forfeit work product protection and that if the plaintiff used generative AI to prepare filings or for hearings, those interactions may be shielded as work product.

During her January 2026 deposition the pro se plaintiff admitted she had an AI platform (ChatGPT) open and was consulting it while answering questions. When asked whether she was feeding information into ChatGPT, she refused to answer citing attorney-client privilege despite having no attorney. The court found the privilege claim inapplicable, instructed that she could not use ChatGPT or any other AI or person to assist her in answering deposition questions, and ordered she must answer questions without relying on AI.

The court held that Federal Rule of Civil Procedure 26(b)(3) can protect a pro se litigant’s use of AI as work product, but on these facts the plaintiff failed to show that identifying the specific AI tool would reveal mental impressions or strategy. The court therefore ordered the pro se plaintiff to disclose the name of any AI platform he used with information designated Confidential, while amending the Protective Order to bar uploading Confidential Information into mainstream AI platforms unless the provider is contractually prohibited from using inputs to train models, discloses inputs only as necessary, and permits deletion of inputs. The ruling protects AI-generated work product in principle but requires tool identification and imposes restrictions on AI use with Confidential material.

(Summary by GPT-5)

Defendants moved to amend the protective order to bar parties from uploading any discovery materials to public/open AI tools, arguing such tools continually train on submitted data (making clawback and deletion impossible), pose data-security and GDPR risks, and could expose critical-infrastructure information. Plaintiffs opposed on burden and First Amendment grounds. The court found good cause, rejecting the characterization as a disfavored umbrella order, concluding the AI-specific risks (inability to claw back data, privacy/GDPR concerns, and critical-infrastructure exposure) justified extending the protective-order restrictions to all discovery while permitting use of secure/closed AI tools.

Defendant Bradley Heppner used the publicly available generative AI platform Claude to create written reports and communications after becoming a target of a grand jury investigation. The court held the documents were neither attorney-client privileged nor protected work product because (1) communications with Claude are not communications with an attorney, (2) Anthropic's privacy policy and retention practices defeated any reasonable expectation of confidentiality, and (3) the materials were not prepared at the direction or behest of counsel and did not reflect counsel's mental impressions or strategy. Accordingly the AI-generated/exchanged documents are not protected.

(Summary by GPT-5)

Defendants sought production of Plaintiff's use of third-party AI tools (e.g., ChatGPT) and argued any work-product protection was waived by such use. The magistrate judge denied the request, holding the information about AI use is not discoverable and is protected by the work-product doctrine (and attorney-client privilege where applicable), finding no waiver from use of AI tools. The court reasoned the requests were irrelevant or disproportional, would invade mental impressions/opinion work product, and that generative AI are tools not persons whose use alone effectuate waiver.

(Summary by GPT-5)

"We also observe that to put client letters and decision letters from the Home Office into an open source AI tool, such as ChatGPT, is to place this information on the internet in the public domain, and thus to breach client confidentiality and waive legal privilege, and thus any regulated legal professional or firm that does so would, in addition to needing to bring this to the attention of their regulator, be advised to consult with the Information Commissioner’s Office. Closed source AI tools which do not place information in the public domain, such as Microsoft Copilot, are available for tasks such as summarising without these risks."