As the use of generative AI (“gen AI”) platforms becomes ubiquitous in both legal practice and business operations, courts are confronting novel questions about whether prompts submitted to—and outputs generated by—platforms such as ChatGPT, Claude, and Microsoft Copilot are discoverable, and whether parties have an obligation to preserve such materials in anticipation of litigation. For IP litigation and general litigation practitioners, these developments carry immediate practical implications: gen AI interaction logs may be relevant, may be subject to work product or privilege protections, and may reside on systems with short default retention periods that are inconsistent with a litigation hold. This alert surveys the emerging case law on discoverability, then focuses in greater depth on when the duty to preserve gen AI materials attaches and on practical, actionable guidance—including recommended hold notice elements, gen AI usage audit steps, and general hold-drafting principles—for preserving gen AI prompts and outputs in anticipation of litigation.

Are Gen AI Prompts and Outputs Discoverable?

Courts have consistently applied the traditional ESI framework under Fed. R. Civ. P. 37(e) to AI prompts and outputs rather than creating a heightened or AI-specific standard. Whether such materials are discoverable depends on relevance, and whether they are shielded from discovery depends on the same privilege and work product doctrines that govern all litigation materials. The cases below illustrate how these principles have been applied.

In re OpenAI, Inc., Copyright Infringement Litig. (S.D.N.Y. 2025) (ChatExplorer Logs – relevance).

In In re OpenAI, Inc., Copyright Infringement Litig., 800 F. Supp. 3d 602 (S.D.N.Y. 2025), OpenAI and Microsoft moved to compel production of all of the New York Times's ChatExplorer Logs—prompts and outputs from NYT's internal tool interfacing with OpenAI's GPT models via API—and sought a four-hour Rule 30(b)(6) deposition on NYT's preservation efforts. The court denied both motions, holding that “[b]ecause the ChatExplorer Logs are not relevant, the inadvertent loss of these logs does not implicate Fed. R. Civ. P. 37(e).” The court cited In re Keurig Green Mountain Single-Serve Coffee Antitrust Litigation, 341 F.R.D. 474, 497 (S.D.N.Y. 2022), which requires a spoliation movant to show that lost information “was not only probative, but that it would affirmatively support the movant's claim.” This decision confirms that gen AI interaction logs are evaluated under the same relevance-driven ESI framework as any other electronically stored information.

Whether gen AI prompts and outputs are protected by attorney-client privilege or the work product doctrine turns on who created them and why. The following cases illustrate the emerging framework.

Tremblay v. OpenAI, Inc., No. 23-CV-03223-AMO, 2024 WL 3748003 (N.D. Cal. Aug. 8, 2024).

The court held that attorney-crafted ChatGPT prompts designed to test legal theories were opinion work product because “the ChatGPT prompts were queries crafted by counsel and contain counsel's mental impressions and opinions about how to interrogate ChatGPT, in an effort to vindicate Plaintiffs' copyrights against the alleged infringements.” Opinion work product “is virtually undiscoverable” and can be compelled “only where mental impressions are at issue in a case and the need for the material is compelling.” Any waiver “must be closely tailored to the needs of the opposing party.”

In re OpenAI, Inc., Copyright Infringement Litig., 802 F. Supp. 3d 688 (S.D.N.Y. 2025) (Oct. 1, 2025) (employee prompt spreadsheet – privilege).

A spreadsheet of employees’ ChatGPT prompts and outputs, created supposedly at in-house counsel's direction, was held NOT privileged because it bore no privileged legend, had no counsel entries, and contained no actual legal advice—“a document will not become privileged simply because an attorney recommended its preparation if it contains merely business-related or technical communications between corporate employees.” In re Rivastigmine Patent Lit., 237 F.R.D. 69, 80 (S.D.N.Y. 2006). This holding means that business or compliance testing records of gen AI use should be expected to be discoverable and must be preserved.

At-Issue Waiver of gen AI Work Product. Courts have also addressed circumstances in which work product protection over gen AI materials is waived by placing them “at issue.”

Concord Music Grp., Inc. v. Anthropic PBC, No. 24-CV-03811-EKL (SVK), 2025 WL 3677935 (N.D. Cal. Dec. 18, 2025).

Work product for post-suit investigation prompts and outputs was waived where the party placed the investigation at issue by indicating investigators and client-witnesses would testify about it. However, pre-suit investigation prompts and outputs were not ordered produced because “Anthropic will be able to conduct an effective cross-examination based on the statistics of Publishers’ pre-suit investigation without resorting to the mental impressions of counsel,” and waiver “must be closely tailored to the needs of the opposing party.”

T.B. v. Big Brothers Big Sisters of New York City, No. 452864/2021, 2025 WL 2443502 (N.Y. Sup. Ct. Aug. 21, 2025).

Counsel’s partial disclosure of a ChatGPT transcript placed it at issue, waiving work product protection. The court ordered production of “the full transcript of the communication with ChatGPT which led to ChatGPT providing the purported summarized article.”

Current Awareness. A federal court has reportedly required production of prompts and queries tied to an expert’s gen AI-assisted workflow as discoverable under Rule 26. See “Produce the Prompts: A Court Says Expert AI Inputs Are Fair Game in Discovery,” 2026 WLNR 15258059. Commentary notes that preservation duties may extend to gen AI processed through cloud deployments such as Microsoft Azure OpenAI, and that failure to preserve may trigger Rule 37(e) consequences. See “When Your Expert Uses AI: The First Federal Decision on Discoverability of Expert Prompts,” 2026 WLNR 18155736. Florida has amended Fla. R. Gen. Prac. & Jud. Admin. 2.515(d)(2)(D) (effective June 15, 2026) to underscore attorney accountability for AI-assisted filings. And the Ninth Circuit's decision in Lnu v. Blanche, 177 F.4th 1014 (9th Cir. 2026), addressing attorneys' ethical duty to rigorously check gen AI outputs for hallucinations, reflects the broader judicial trend of imposing affirmative obligations regarding gen AI use in litigation.

When Does the Duty to Preserve AI Materials Attach?

The duty to preserve electronically stored information—including AI prompts and outputs—arises from the common-law obligation to preserve relevant evidence when litigation is “reasonably anticipated.” Fed. R. Civ. P. 37(e) “is based on this common-law duty; it does not attempt to create a new duty to preserve.” 2015 Advisory Committee Notes to Rule 37(e). Anticipation of litigation can arise well before a complaint is filed: the duty is triggered once a company identifies a potential claim, receives a complaint or notice of lawsuit, receives a third-party subpoena, receives a formal investigation order, or becomes or should become aware of a potential legal claim. See Implementing a Litigation Hold, Practical Law Practice Note 8-502-9481.

No court has created a heightened or gen AI-specific preservation standard. Courts apply the same relevance and reasonableness test used for all ESI. See In re OpenAI, Inc., Copyright Infringement Litig., 800 F. Supp. 3d 602 (S.D.N.Y. 2025). The practical challenge lies in the fact that gen AI interaction logs often reside on third-party platforms with short default retention periods—a mismatch that can quickly convert a good-faith oversight into a spoliation problem.

Because gen AI interaction data is often ephemeral, dispersed across multiple platforms, and subject to short default retention windows, the preservation obligations described below take on heightened practical significance in the gen AI context. The following authorities—addressing counsel’s continuing preservation obligations, the consequences of failing to issue a timely hold, and the specific challenges posed by gen AI logs—illustrate how these general principles apply.

Counsel’s continuing obligation. The issuance of a litigation hold does not end counsel’s preservation duties. Counsel “must continue to monitor and supervise or participate in a party’s efforts to comply with the duty to preserve.” DR Distributors, LLC v. 21 Century Smoking, Inc., 513 F. Supp. 3d 839 (N.D. Ill. 2021). After receiving a litigation hold, “an individual must take steps to determine what is necessary to comply. That obligation includes learning of systems ‘to prevent destruction or automatic deletion.’” Goldstein v. Denner, 310 A.3d 548 (Del. Ch. 2024). “For an organization, simply circulating a litigation hold is not enough. The organization must take steps to ensure that the recipients of the hold understand what it means and abide by it. The organization also must suspend or modify routine document retention or document destruction policies so that evidence is not lost.” Id.

Consequences of failing to issue a hold. A party that fails to take reasonable steps to preserve relevant evidence once litigation is reasonably anticipated may be sanctioned for spoliation. In Zubalake v. UBS Warburg LLC, 220 F.R.D. 212 (S.D.N.Y. 2003) (Zubulake IV), the court explained that “[o]nce a party reasonably anticipates litigation, it must suspend its routine document retention/destruction policy and put in place a ”litigation hold” to ensure the preservation of relevant documents”. Id. at 218. The duty is broad: “anyone who anticipates being a party or is a party to a lawsuit must not destroy unique, relevant evidence that might be useful to an adversary”, and a litigant “is under a duty to preserve what it knows, or reasonably should know, is relevant in the action, is reasonably calculated to lead to the discovery of admissible evidence, is reasonably likely to be requested during discovery and/or is the subject of a pending discovery request.” Id. at 217. Because “a party can only be sanctioned for destroying evidence if it had a duty to preserve it”, id at 216, the failure to implement a hold that causes the loss of relevant materials exposes the party to sanctions. In a companion ruling, the court imposed sanctions where “UBS failed to preserve relevant e-mails, even after receiving adequate warnings from counsel.” Zubulake v. UBS Warburg LLC, 229 F.R.D. 422, 439 (S.D.N.Y. 2004) (Zubulake V).

How the duty plays out for gen AI logs: the ChatExplorer Logs. The In re OpenAI litigation provides an instructive example. After in-house counsel directed a hold on the ChatExplorer Logs, an NYT employee tasked with implementing the hold mistakenly believed the logs were already under a one-year retention policy. The logs were actually stored in a SumoLogic partition with only a 90-day retention period. When the error was discovered, the Times migrated the logs to a partition with a 360-day retention policy and simultaneously wrote the data to Amazon S3 with a 10-year retention policy. This illustrates that gen AI interaction logs may reside on short default retention schedules that are inconsistent with a litigation hold, and that a custodian’s assumptions about retention policies cannot be taken at face value.

Public gen AI platform terms of service and data retention and disclosure policies can affect both the confidentiality of gen AI materials and the practical ability to preserve them. Because prompts and outputs may reside on the gen AI provider’s own servers rather than solely on the organization’s systems, counsel issuing a hold must account for this additional layer of preservation complexity, including confirming whether the organization can direct the platform to suspend automatic deletion or must instead export and retain the data independently. This is particularly important given that, as discussed above, the duty to preserve under Fed. R. Civ. P. 37(e) applies with equal force to gen AI interaction logs regardless of where they are hosted.

Best Practices for Drafting Litigation Holds Covering Generative AI

A written litigation hold letter remains the standard and recognized method for documenting preservation obligations. “Best practices dictate that a written letter is the superior form of the litigation hold notice for various practical reasons, primarily because it documents when the hold issues and what it directs a litigant to do to meet its discovery obligations.” DR Distributors, LLC v. 21 Century Smoking, Inc., 513 F. Supp. 3d 839 (N.D. Ill. 2021) (quoting Borum v. Brentwood Vill., LLC, 332 F.R.D. 38, 45 (D.D.C. 2019)). Simply directing a client to “save everything” is insufficient; “attorneys must give reasonable and specific instructions detailing where ESI might be stored and what steps the client may need to take to preserve it.” Id. An adequate hold notice must include a warning to disable autodelete functions. Id.

No rule or standing order yet mandates specific language for litigation holds addressing gen AI. However, leading practice guidance has emerged to fill the gap. The following elements should be included in any gen AI-inclusive hold notice:

• Definition of Covered gen AI Materials. Expressly define the scope of the hold to include: (a) all prompts, queries, or instructions submitted to any generative AI tool or platform, including but not limited to ChatGPT, Claude, Microsoft Copilot, Google Gemini, DeepSeek, Grok, and any proprietary or enterprise gen AI tools; (b) all outputs, responses, or content generated by any such tool in response to such prompts; (c) all logs, records, or metadata associated with such gen AI interactions; and (d) any documents or records that describe, evaluate, or summarize gen AI prompts or outputs, including spreadsheets, notes, or reports created in connection with gen AI testing or use.

• Identification of Storage Locations. Direct recipients to identify all locations where gen AI interaction logs may be stored, including: (a) the gen AI platform's own servers or cloud storage (and the applicable retention period for such storage); (b) any internal systems, databases, or logs that capture gen AI usage; (c) local devices, including computers, tablets, and mobile phones; and (d) any third-party storage services used in connection with gen AI tools.

• Suspension of Automatic Deletion. Instruct recipients to suspend any automatic deletion, purging, or overwriting of gen AI interaction logs, and to contact IT personnel to ensure that default retention settings on gen AI platforms and associated storage systems are overridden to prevent the loss of potentially relevant materials.

• Distinguish Open vs. Closed gen AI Tools. For open (public) platforms, instruct recipients to export or download interaction logs before they are automatically deleted by the platform. For closed (proprietary/enterprise) platforms, instruct recipients to ensure the platform's settings are configured to retain interaction logs for the duration of the litigation hold.

• Confidentiality Caution. Caution recipients against submitting any documents or information subject to the hold into any open gen AI tool, as doing so may constitute disclosure to a third party and could compromise the confidentiality of privileged or sensitive materials, particularly where the platform’s privacy policy permits collection or disclosure of user inputs and outputs to third parties.

• Periodic Reissuance and Reminders. Because counsel’s preservation obligations continue throughout the litigation, hold notices addressing gen AI materials should be reissued or supplemented as new gen AI platforms are adopted, and recipients should be periodically reminded of their ongoing obligation to preserve gen AI interaction logs and to avoid enabling auto-deletion on new tools.

Conduct a Gen AI Usage Audit. As part of the hold process, and consistent with counsel’s continuing obligation to monitor compliance, organizations should conduct a thorough gen AI usage audit to ensure that all relevant sources of AI-generated evidence are identified and preserved:

• Identify all gen AI platforms used by employees relevant to the litigation's subject matter.

• Determine each platform's default log retention period.

• Take affirmative steps to extend retention or export logs before automatic deletion.

• Document all preservation steps taken.

• Reassess gen AI usage and retention settings periodically throughout the litigation, since new platforms may be adopted or existing default settings may change over time.

General Hold Principles. Several principles apply to any litigation hold, with or without a gen AI component:

• The duty to preserve is triggered by reasonably anticipated litigation, which can precede a filed complaint.

• A hold notice—written or oral—must contain enough information to adequately inform the recipient of discovery obligations under the circumstances. “The preservation obligation runs first to counsel, who has a duty to advise his client of the type of information potentially relevant to the lawsuit and of the necessity of preventing its destruction.” Orbit One Commc'ns. v. Numerex Corp., 271 F.R.D. 429, 437 (S.D.N.Y. 2010).

• Issuing the hold is not the end of counsel's obligation—counsel must monitor and supervise compliance. DR Distributors, 513 F. Supp. 3d at 839.

• Preservation obligations apply regardless of where the relevant data resides, including on third-party or vendor-hosted systems such as cloud-based gen AI platforms, and counsel should confirm the organization’s practical ability to control retention on such systems before representing that a hold has been fully implemented.

Key Takeaways

Given the rapidly developing case law, practitioners should take the following steps now:

• Apply the existing ESI framework. Courts treat gen AI prompts and outputs as electronically stored information subject to the same discovery and preservation rules as all other ESI. Do not assume that gen AI-related materials require—or benefit from—a separate standard.

• Update litigation hold templates. Revise hold notices to expressly define and cover gen AI prompts, outputs, interaction logs, metadata, and any documents summarizing or evaluating gen AI use.

• Identify and map gen AI storage locations. Instruct hold recipients and IT to identify all platforms where gen AI interaction data resides, including third-party cloud storage with short default retention periods.

• Suspend auto-deletion immediately. Ensure that default retention settings on gen AI platforms are overridden as part of hold implementation, and export or preserve logs from public platforms before they are automatically purged.

• Protect privilege by controlling the process. Work product protection is most likely to attach when AI prompts and outputs are prepared at counsel's direction and reflect counsel’s mental impressions. Materials created independently by a client without counsel’s involvement—particularly on public platforms with broad disclosure policies—may not be protectable.

• Anticipate at-issue waiver. If witnesses will testify about a gen AI-assisted investigation, the underlying prompts and outputs may become discoverable. Account for this risk when deciding whether to disclose gen AI use in litigation.

• Monitor and supervise compliance. A hold letter alone is insufficient. Counsel must follow up with custodians, verify that auto-deletion has been disabled, and confirm that gen AI interaction logs are being preserved in a manner consistent with the hold's instructions.

Courts continue to apply traditional ESI and privilege/work-product doctrines to gen AI materials without creating a separate standard, and this area of law is developing quickly. 

We will continue to monitor the emerging case law and guidance from courts regarding the discoverability and preservation of AI-generated content. Please contact us if you have questions about how these decisions may affect your organization.