New Technology, New Discovery

People love playing with new technology. There is no shortage of examples of people asking about any of the popular GenAI applications for everything from how to bake a pie, develop a workout routine, or draft arguments for a lawsuit.  

Here is the problem: the technology people use can become central evidence in a lawsuit. The big question is: Does using GenAI mean that its data is discoverable in a lawsuit? The answer: if the data produced by GenAI is relevant to the lawsuit, yes.  

In federal lawsuits, parties can seek records on any nonprivileged matter that is relevant to any party's claim or defense and proportional to the needs of the case. State laws often have different language relevancy, but the message is consistent: any data sought in discovery must be relevant to the lawsuit.  

In the early days of social media, there were often fishing expeditions seeking every social media post by an adverse party. Such discovery requests were often deemed to be overly broad for data that was not relevant to a lawsuit. It is one thing to request a social media post about mechanical bull riding in a workers’ comp case, but such videos would have no relevance in a breach of contract case.  

The same standards appear to apply to the use of GenAI tools. No one should be conducting fishing expeditions, but if the data is relevant, it is fair game in a lawsuit.  

What are the types of GenAI potentially subject to discovery?

There are many types of GenAI applications. Here are some examples:  

Large language models (LLMs), such as ChatGPT, Claude and Google Gemini, are the type that most people are familiar with from popular usage. These LLMs have also notoriously starred in cases of lawyers and pro se litigants getting into serious trouble with hallucinated case citations.¹  

Generative Adversarial Networks (GANs) can create pictures, video, text, and sound. GANs are effectively two algorithms competing against each other to “create” realistic content.²  

Diffusion models are used for creating images and videos, leveraging a process called “iterative denoising.” The media is created by user prompts.³  

Neural Radiance Fields (NeRFs) create 3D images with deep learning.

The 4 Types Of Generative AI Transforming Our World, Forbes, Bernard Marr, Apr 29, 2024.

The business uses for GenAI applications include software development, content development, financial modeling, healthcare, and more. Consider medical appointments where doctors use an AI note taker. Or if you have been on a video call recently, where every attendant also had an AI bot, join the call to take notes and provide a summary.

GenAI Apps in Litigation

Lawsuits with GenAI are working their way through the courts. One recent case involved allegations that Meta materially contributed to fraudulent ad development by offering generative AI tools that generated the allegedly fraudulent content. Bouck v. Meta Platforms, Inc., No. 25-cv-05194-RS, 2026 LX 189150 (N.D. Cal. Mar. 24, 2026).  

The lawsuit claimed that advertisements were optimized with “creative enhancements” using a GenAI application. The enhancements included AI-generated text and images. Bouck, at *8-9.  

The application could alter contents to improve performance of the ads. Id.

The allegations in the complaint claimed Meta went beyond offering tools to being a “conspirator” in the creation of the fraudulent content. Suddeth v. Meta Platforms, Inc., No. 25-cv-08581-RS, 2026 U.S. Dist. LEXIS 62629, at *9 (N.D. Cal. Mar. 24, 2026), citing Bouck v. Meta Platforms, Inc., No. 25-cv-05194-RS, 2026 LX 189150 (N.D. Cal. Mar. 24, 2026). This means that the parties likely will focus the discovery requests on the AI-application that generated the text and images on the allegedly fraudulent content. This would likely include the prompts to create the ads. Moreover, the discovery requests could focus on how the GenAI works to demonstrate there was indeed a conspiracy between the end-user and the defendant.  

There are many questions about whether the plaintiffs correctly describe the GenAI application. The biggest issue is whether a party that creates a GenAI application can be a conspirator just by making a product for the market.  That question likely will be the subject of discovery requests and depositions to answer if the defendant participated in the ultimate creation of the fraudulent ads or merely had a neutral GenAI tool used by scammers to commit fraud.  

Requests for Production of Training Data

There are multiple lawsuits concerning intellectual property and copyright infringement claims regarding GenAI training data. In the Mosasic LLM Litigation, the plaintiffs served Microsoft with a subpoena for all licensing agreements for training data and licensing books for use as training data. In re Mosaic LLM Litig., No. 24-cv-01451-CRB (LJC), 2025 U.S. Dist. LEXIS 153961, at *2 (N.D. Cal. July 8, 2025).  

Microsoft produced all the licensing agreements for the training data and books, but a discovery dispute followed over other communications.¹⁰ Id.

Sealing Order for User Prompts

Litigation with GenAI companies raises complex issues of third-party data that can contain personal identifiable information, confidential business information, and technical information protected by trade secret laws. Courts are beginning to address these issues with protective orders and sealing orders.  

In one recent case, Anthropic sought a court order to seal user prompts that contained direct quotes from Claude users. Concord Music Grp., Inc. v. Anthropic PBC, No. 24-cv-03811-EKL (SVK), 2025 U.S. Dist. LEXIS 99065, at *11-12 (N.D. Cal. May 23, 2025). The court denied the motion to seal. The court reasoned that the protective order already in place allowed Anthropic to designate prompts as confidential but required Anthropic "de-designate and/or propose tailored redactions for th[e] specific prompt/output record[s]" that it submits in filings. Id. While the court recognized that users had a privacy interest in their prompts, the court found that there is no cause to seal the descriptions and anonymous excerpts. Id.  

The Google Generative AI Copyright Litigation reached a different result on granting a sealing motion. In that case, the parties brought an omnibus sealing motion for documents that included contracts with third parties, business strategy and negotiations, pricing and market analysis, technical research and development, GenAI model training, dataset composition, data acquisition methods, and data indexing and tracking.  

The court did not list all of the reasons for granting the sealing order, because the justification contained confidential information. However, the court noted that the data contained personally identifiable information that was not relevant to a class certification motion. In re Google Generative AI Copyright Litig., No. 23-cv-03440-EKL, 2026 U.S. Dist. LEXIS 50738, at *9 (N.D. Cal. Mar. 11, 2026).

GenAI is a new technology, but the method for protecting confidential information is still the same. Parties can seek a protective order, and if necessary, a sealing order under specific conditions to ensure PII, business information, and trade secrets that are used as inputs for GenAI or are included in the content it generates are protected.  

When Gen AI could be Privileged

Privilege analysis is one of the foundations of document review. No lawyer wants to explain to a client how a privilege was waived.  

The test for determining whether communication or information is privileged turns on the type of privilege at issue. Let’s explore the attorney-client privilege.  

The attorney-client privilege is defined as a confidential communication between a lawyer and client regarding legal advice. Cal. Evid. Code § 954. Evidence codes across the United States vary, but the core elements of the privilege are that an attorney is either being asked for legal advice by a client or providing legal advice to a client.  

There are multiple permutations of the attorney-client privilege. A corporate client can have subsequent internal communication between executives on implementing legal advice from counsel that is protected by the attorney-client privilege. Zurich Am. Ins. Co. v. Superior Court, 155 Cal. App. 4th 1485, 1497, (2007). Additionally, communications exchanged between the attorney, client, and insurance carrier regarding legal advice are also protected under privilege. Heffron v. L.A. Transit Lines, 170 Cal. App. 2d 709, 718 (1959).

Consider the following: An attorney provides litigation hold instructions for preserving evidence in a lawsuit. The client then uses an AI application to identify data sources that might have relevant information. In that instance, the client is using an AI application at the direction of an attorney, enacting legal advice to preserve relevant information. Those AI prompts would likely be considered protected by the attorney-client privilege.  

GenAI privilege issues appeared in two orders in federal courts in New York and Colorado with seemingly different results on whether the use of GenAI applications can be protected by the work product doctrine.  

In the New York criminal case, District Judge Jed S. Rakoff in the U.S. District Court for the Southern District of New York held that the work product doctrine did not apply to the criminal defendant who used a GenAI application without the knowledge or direction of his attorney. United States v. Heppner, 2026 WL 436479 (S.D.N.Y. Feb. 17, 2026).

Judge Rakoff applied a three-part test to determine if the records were protected by the attorney-client privilege:

1. Was the communication between a client and attorney?

2. Were the communications kept confidential?

3. Were the communications prepared for obtaining legal advice?

The court held the attorney-client privilege did not apply to the GenAI information, because there were no communications between an attorney and client about legal advice. Simply put, a GenAI application is not a lawyer, let alone a human being. Judge Rakoff focused on Claude’s terms of use to find the prompts made to Claude were not confidential. The court explained that Anthropic user prompts to train its GenAI, thus per the terms of service; an end user does not have an expectation of privacy.  Furthermore, the court held that Claude was not used for seeking legal advice. The issue for the court was that the defendant used a GenAI application without the direction of counsel.  

The court also held that the work product doctrine did not protect the Claude data, because the defendant was not acting under the direction of counsel.  

In the Colorado civil action, Magistrate Judge Maritza Dominguez Braswell in the U.S. District Court District of Colorado held that the work product doctrine applied to a pro se litigant using a GenAI application.  Judge Braswell explained that Federal Rule of Civil Procedure 26(b)(3) applies to material 1) prepared by a party before retaining a lawyer and 2) a party who never actually hires an attorney. This included a pro se litigant representing themselves in a civil lawsuit. Morgan v. V2X, Inc., Civil Action No. 25-cv-01991-SKC-MDB, 2026 U.S. Dist. LEXIS 67939 (D. Colo. Mar. 30, 2026).

The civil procedure ruling is significant, but there is a larger issue about the technology that was decided in Morgan: "AI interactions do not automatically compromise work product protections."

Judge Braswell explained that virtually all data passes through a third-party system, from smartphones to search engines. Case law has held that email subscribers have a reasonable expectation of privacy of their emails stored by internet service providers. United States v. Warshak, 631 F.3d 266, 268 (6th Cir. 2010). Furthermore, the U.S. Supreme Court has held that a person's reasonable expectation of privacy in data is not automatically extinguished by the data being held by a third-party intermediary. Carpenter v. United States, 585 U.S. 296, 310-16 (2018).

Based on the above, the court explained that there is an arguably stronger privacy argument in the context of modern AI because the applications are designed to engage with the end user. The features include simulating empathy and endearing trust that feels genuine. Morgan, at *12-13.

The argument that if an attorney, or expert witness at the direction of an attorney, uses a GenAI, means that privileged information was compromised, does not mean that a privileged was waived under Morgan.

There will be other nuanced court orders in the months ahead about when an individual’s use of GenAI in criminal or civil proceedings will be protected by the work product doctrine. It is likely that whether litigation was anticipated or not will be key in determining if the work product doctrine applies in civil lawsuits. As for criminal cases, there could be situations where a defendant is representing themselves where different privileges apply.  

Courts Taking the Initiative on GenAI

Judges across the U.S. are taking action on understanding uses of GenAI. There are 1300 court orders ruling on GenAI hallucinated case citations that run the gamut from warnings to sanctions for citing nonexistent case law. Damien Charlotin, AI Hallucination Cases https://www.damiencharlotin.com/hallucinations/, Last Visited 04/24/2026.  

That level of frequency cannot be ignored. In Washington D.C., the District of Columbia Courts appointed a task force with goals that include, “Develop[ing] court rules with respect to disclosure, transparency, ethics, accuracy, authenticity and certification of AI use in court pleadings and proceedings.” In re A.I. Task Force of the D.C. Courts, 2026 D.C. App. LEXIS 82, at *4-5 (Mar. 6, 2026).

The AI task force goals of accuracy and certification connect directly to discovery and motion practice. Attorneys argue about facts and their application to the law before judges every day. That information must be right to meet a lawyer’s ethical duty of candor to the court. How information is collected, analyzed, produced to opposing parties, and used in arguments, cannot simply be, “the GenAI said so.” Attorneys need to be able to explain how they used a GenAI application in a defensible manner.  

Courts will provide leadership on the use of GenAI by attorneys, whether it is from a task force or court orders on the use of GenAI. State bar associations have already promulgated ethical rules on the use of GenAI. California is now considering an amendment on ethical rules for what GenAI a lawyer must “independently review, verify, and exercise professional judgment regarding any output generated by the technology.” Proposed Amendments to the Rules of Professional Conduct Related to Artificial Intelligence, https://www.calbar.ca.gov/public/public-meetings-comment/public-comment/public-comment-archives/2026-public-comment/proposed-amendments-rules-professional-conduct-related-artificial-intelligence, Last visited on 04/24/2026.  

GenAI has the promise of helping lawyers practice law more effectively. However, the number of times judges have warned, or sanctioned lawyers, highlights the need for attorneys to be competent in their use of GenAI.  

This is the Beginning of GenAI in Discovery

We are just at dawn of this new technology. Where GenAI ultimately goes is not yet known. What is known is that people are using this technology to work more efficiently and if that usage is relevant to a lawsuit, it should be preserved, collected, and produced in litigation.  

‍