Will Courts Enforce Protective Order Restrictions Affecting AI Use?
Editor’s Note: Protective orders have historically been one of the more routine aspects of discovery practice, with disputes typically confined to narrow questions of access and designation. Generative AI (GenAI) has introduced a new and largely unresolved issue: how courts will treat the unauthorized use of confidential litigation material in AI tools. As contributing author Phil Favro explained in this article, no court has yet ruled directly on this question. The closest available guidance comes from a magistrate judge who indicated that future violations could prompt a show-cause order, along with two recent MDL rulings—Uber and Acetaminophen—in which courts sanctioned attorneys for misusing protected information in other, non-AI contexts. Those sanctions were real but relatively modest, which raises questions about their deterrent value as AI-related violations become more common. Practitioners should consider using this opportunity to address AI use explicitly in protective order negotiations rather than wait for courts to set the standard.
Will Courts Enforce Protective Order Restrictions Affecting AI Use?
By Phil Favro, Contributing Author for HaystackID
Protective order practices have been upended by artificial intelligence (AI).
Until the advent of AI, protective orders had long been standardized. Many courts had model protective orders to which parties often agreed and which courts entered as a matter of course. To be sure, parties might squabble over who should have access to “attorney’s eyes only” (AEO) documents. And there have always been fights about whether certain documents should even be designated “confidential.” These were mere skirmishes, though, and typically did not predominate discovery practice issues.
AI has changed this. Parties now must consider whether and to what extent protective orders should restrict the use of designated materials in AI applications. The purpose of these restrictive provisions is to prevent confidential, proprietary, or other sensitive information[1] from being ingested into AI applications for training or improving AI models. For parties attempting to safeguard such information, provisions that limit its use in AI applications are viewed as a bulwark against its misappropriation in litigation and the possible loss of confidentiality.
Sanctions for Protective Order Violations
But what happens if parties or their lawyers violate those provisions by uploading confidential information into an unauthorized AI application? Will courts impose sanctions for protective order violations?
And if courts do levy sanctions, will the punishments be sufficiently severe to deter future violations, either in the instant case or in other lawsuits?
Courts have not directly addressed these questions. At least one court—Campbell v. TidalHealth, Inc.—deferred when confronted with the possibility of such a violation. The Campbell court (Magistrate Judge Charles Austin) suggested it would “be more inclined to issue a show cause order and consider sanctions” should “future filings suggest improper submission of discovery materials to generative artificial intelligence tools.”[2]
The lack of any definitive action in Campbell regarding alleged misuse of protected materials in AI tools should concern organizations. And yet, courts have imposed sanctions in recent cases where protective order violations have transpired, albeit in different contexts. In In re Uber Technologies, Inc., Passenger Sexual Assault Litigation and In re Acetaminophen – ASD-ADHD Products Liability Litigation, courts issued monetary sanctions and other penalties after determining that attorneys misused protected information by disclosing it in other cases.
Uber and Acetaminophen confirm that courts will consider sanctions for protective order violations. However, those sanctions may not adequately remediate harm arising from unauthorized disclosures of confidential information in AI applications, depending on the content of the exposed information. If anything, these cases spotlight the need for parties to consider additional approaches in protective orders to address unauthorized uses of confidential information in AI tools.
Uber—Monetary and Other Sanctions
The protective order infractions from the Uber case—a multidistrict litigation (MDL) arising from instances of sexual misconduct by drivers for the defendant’s (Uber) car service—involved the disclosure of Uber’s confidential information from that MDL in other lawsuits.[3] A lawyer representing the plaintiffs (Stanley) was responsible for those disclosures, which the court found resulted in multiple violations of the protective order and warranted the imposition of sanctions.
The First Violation
The first violation of the protective order transpired when Stanley revealed to attorneys in other lawsuits against Uber “the complete substantive contents of certain documents produced in discovery and designated as confidential, i.e., lists of policy documents and the manner in which Uber organized them.” To address this violation, the parties settled on a proposed order with measures to remedy harm caused by the disclosure (hereafter the “August 18, 2025 Order”). Among other things, Stanley agreed to identify the court cases where he had used or disclosed Uber’s confidential information, along with the unauthorized persons who received that information. In addition, Stanley was to provide a copy of the stipulated order to the courts and persons he identified and make “reasonable efforts” to destroy the confidential information he improperly shared with others.
After Stanley apparently did not comply with certain aspects of the August 18, 2025 Order, Uber sought judicial relief in the MDL. The court (Magistrate Judge Lisa Cisneros) determined that Stanley violated both the protective order and the August 18, 2025 Order. According to Judge Cisneros, violations of those orders were actionable under Federal Rule of Civil Procedure 37(b)(2). And because Stanley’s actions were not substantially justified, monetary sanctions should issue pursuant to Rule 37(b)(2)(C).
Yet Judge Cisneros awarded Uber just $30,000 in sanctions, far below the $168,572 Uber sought. The judge provided several reasons why she granted a reduced fee award. Among those were Uber’s litigation conduct (apparently delaying its request for sanctions against Stanley) and its apparent inability to substantiate its fee request. Judge Cisneros also suggested that Stanley’s disclosure of confidential information was not unduly harmful to Uber given the nature of his disclosure (sharing it with other lawyers and courts rather than with the media).
Moreover, Judge Cisneros felt that this type of violation was not significant. While emphasizing that she was not condoning Stanley’s actions, Judge Cisneros reasoned that Uber’s interest in preventing Stanley from using its confidential documents to pursue claims against the company on behalf of different clients in other lawsuits “carries at most limited weight.” The judge reasoned that prevailing circuit case law favors allowing access to discovery materials in collateral litigation to advance notions of judicial economy.
The Second Violation
Stanley’s second protective order violation transpired when he identified a confidential-designated document in the MDL and then unilaterally used information from that document in a Texas-based lawsuit against Uber involving another of his clients. Judge Cisneros again determined this to be a straightforward violation of the MDL protective order and authorized the imposition of monetary sanctions.
After negotiations, the parties stipulated that Stanley would pay Uber $50,000 to compensate the company for its attorney’s fees and costs incurred in bringing the second motion for sanctions.[4] As an additional sanction, the district judge (Honorable Charles Breyer) authorized Stanley’s immediate removal from the Plaintiffs Steering Committee in the MDL.[5]
Acetaminophen—Monetary Sanctions
Monetary sanctions were also imposed on a lawyer in the Acetaminophen MDL after a court found multiple protective order violations.[6] Acetaminophen involves claims that “children developed autism spectrum disorder and/or attention-deficit/hyperactivity disorder as a result of in utero exposure to acetaminophen.”
Like Uber, the violations in Acetaminophen involved an attorney (Keller) disclosing or using confidential information from that case in other lawsuits against the same defendant (Kenvue). In each of those other lawsuits (respectively venued in Florida and Texas), Keller and his firm (Keller Postman) apparently used or disclosed confidential information from the Acetaminophen MDL to substantiate their opposition briefs to Kenvue’s motions to dismiss for lack of personal jurisdiction.
In the Florida action, Keller and his team of lawyers took action to strike the confidential information from their opposition papers in response to Kenvue’s assertion that Keller improperly used that information in violation of the MDL protective order. However, Keller and other attorneys representing their clients in the Texas litigation apparently took no similar action to address their apparent violations of the MDL protective order. Instead, they attempted to file unredacted copies of their opposition papers under seal in the Texas litigation, asserting that discovery from the Acetaminophen MDL demonstrated that Kenvue was subject to personal jurisdiction for purposes of the Texas case.
The Acetaminophen court (District Judge Denise Cote) found that Keller’s actions and those of his firm constituted violations of the MDL protective order and warranted the imposition of monetary sanctions.[7] In so doing, Judge Cote distinguished Keller’s actions in the Texas case from those in the Florida case.
In the Florida litigation, Keller and his legal team immediately withdrew opposition papers after being apprised of the protective order violation and “filed a revised motion omitting any reference to the confidential information.” According to Judge Cote, Keller’s “prompt corrective action” belied any notion of willfulness that might have merited sanctions. In contrast, the judge reasoned that Keller and his firm persisted in filing opposition papers in the Texas litigation that used or referenced confidential information over the objections of Kenvue. Moreover, Judge Cote observed that this violation, which began in or about February 2026, remained ongoing at the time she issued her order on May 8, 2026.
Judge Cote ordered Keller and his law firm to pay Kenvue’s reasonable attorney’s fees, which the parties stipulated would be satisfied by a $50,000 donation to a charitable organization.[8] The court also directed that Keller require every lawyer at his firm “working on the [Acetaminophen] MDL, any Related Action, and any action in which Keller Postman seeks to rely on MDL protected discovery material to read [the court’s order] and . . . file an affidavit within two weeks confirming that he has done so.”
Possible Safeguards for Protective Orders
Uber and Acetaminophen demonstrate that courts may impose sanctions for protective order violations. Nevertheless, it is still an open question whether those sanctions will actually deter future protective order infractions. Indeed, it appears the lawyers who violated the protective orders in those cases were not particularly repentant about their conduct. In Uber, Stanley’s second protective order violation was both “clear-cut” and “occurred after the Court had already found previous violations as discussed above.” In Acetaminophen, despite the court’s finding of willfulness, Keller still maintains that he did nothing wrong and has even filed an appeal from the court’s order.
While Uber and Acetaminophen did not involve protective orders with AI-related restrictions, these cases could be instructive on whether sanctions might deter parties or their counsel from misusing confidential materials and improperly ingesting them into an AI application. If Uber and Acetaminophen are exemplary, then monetary sanctions may not sufficiently disincentivize parties or their counsel from misappropriating confidential information. And yet, Stanley’s removal from the Plaintiffs Steering Committee clearly establishes the Uber court’s escalating concern with his protective order violations. In any event, there is no guarantee—as Campbell shows—that courts will even take action to address protective order violations of AI-restrictive provisions.
Given the possible low deterrence value of sanctions and that a court may even defer ruling on a protective order violation, Uber, Acetaminophen, and Campbell spotlight the need for counsel to consider proactive approaches in protective orders as they relate to AI-restrictive provisions. The particular approach to be adopted will likely depend on the type of case at issue, the nature and extent of the confidential information involved, and the manner or methods of misuse of confidential information.
For example, symmetrical litigation matters—in which each party has confidential information they wish to safeguard from misuse in AI applications—should offer counsel an opportunity to negotiate and include in the protective order the names or types of AI applications where parties are authorized to disclose particular types of confidential information. Counsel could also consider including one or a combination of graded and progressive remedies for unauthorized disclosures of such information. For example, the parties could delineate more severe sanctions for misusing AEO materials than for just confidential information. Or sanctions could become progressively more severe with each violation of a protective order.
This approach may or may not be workable in other cases. In matters involving pro se litigants or asymmetrical litigation, individual parties may not have the same need or incentive to safeguard confidential information as corporate parties. Nevertheless, even failed negotiations with adversaries over remedies to address violations of AI-restrictive provisions could still influence a court, either near-term or further in the future, regarding any sanctions that could be issued to address protective order infractions. Indeed, one of the factors mitigating against a more substantial sanctions award in Uber was the lack of notice to Stanley regarding Uber’s intent to seek sanctions for protective order violations (“Uber appears to have sandbagged its request for sanctions for violation of the Protective Order.”). By emphasizing the importance of maintaining confidentiality, identifying the harm that could arise from divulging confidential information in unauthorized AI applications, and specifying possible sanctions during protective order negotiations, litigants could avoid the charge of “sandbagging” and spotlight the legitimate nature of the requested sanction.
For the time being, there isn’t a foolproof way for litigants to completely prevent adversaries from revealing confidential information to unauthorized AI tools. Mistakes will happen, and sometimes disclosures may be intentional rather than inadvertent. Making some effort to consider appropriate remedies for these violations is essential if litigants wish to better safeguard confidential information in the AI era.
[1] Confidential, proprietary, private, or other sensitive information is generally referred to in this article as “confidential information.” [2] Campbell v. TidalHealth, Inc., No. 1:25-CV-4293-BAH, 2026 WL 2162749 (D. Md. July 28, 2026). [3] In re Uber Techs., Inc., Passenger Sexual Assault Litig., No. 23-MD-03084-CRB (LJC), 2026 WL 440752 (N.D. Cal. Feb. 17, 2026). [4] In re Uber Techs., Inc., Passenger Sexual Assault Litig., No. 23-MD-03084-CRB (LJC), ECF No. 5644 (Amended Stipulation and Order regarding Uber Technologies, Inc., Rasier, LLC, Rasier-CA, LLC’s Second Motion for Sanctions against Bret Stanley) (N.D. Cal. Mar. 25, 2026). [5] In re Uber Techs., Inc., Passenger Sexual Assault Litig., No. 23-MD-03084-CRB (LJC), ECF No. 5268 (Order regarding Motions for Sanctions against Bret Stanley) (N.D. Cal. Feb. 18, 2026). [6] In re Acetaminophen – ASD-ADHD Prods. Liab. Litig., No. 22MD3043 (DLC), 2026 WL 1266003 (S.D.N.Y. May 8, 2026). [7] Judge Cote found that Keller’s actions separately violated the court’s “Coordination Order.” The Coordination Order was essentially an enabling order that both allowed parties to use confidential materials from Acetaminophen in other litigation beyond the MDL and “described the steps that would ensure that those litigants adhered to the confidentiality provisions in the Protective Order.” [8] See In re Acetaminophen – ASD-ADHD Prods. Liab. Litig., No. 22MD3043 (DLC), ECF No. 1657 (Order) (S.D.N.Y. June 3, 2026).
About Phil Favro
Phil Favro is the founder of Favro Law PLLC, where he counsels clients on ESI, AI, and discovery issues and serves as a special master, mediator, and expert witness. Phil is nationally recognized for his expertise on ESI, discovery, and information governance, with courts acknowledging his credentials. See, e.g., Oakley v. MSG Networks, Inc., No. 17-CV-6903 (RJS), 2025 WL 2061665 (S.D.N.Y. July 23, 2025). This background makes Phil particularly well-suited to counsel clients and advise courts on information-related issues. As a special master, Phil is acclaimed for his collaborative approach, working with parties to find stipulated solutions to complex issues. For disputes that require adjudication, he is renowned for the clarity and vigor of his written dispositions, which are available on legal search engines.
About HaystackID®
HaystackID® solves complex data challenges related to legal, compliance, regulatory, and cyber requirements. Core offerings include Global Advisory, Cybersecurity, Core Intelligence AI™, and ReviewRight® Global Managed Review, supported by its unified CoreFlex™ service interface and eDiscovery AI® technology. Recognized globally by industry leaders, including Chambers, Gartner, IDC, and Legaltech News, HaystackID helps corporations and legal practices manage data gravity, where information demands action, and workflow gravity, where critical requirements demand coordinated expertise, delivering innovative solutions with a continual focus on security, privacy, and integrity. Learn more at HaystackID.com.
Assisted by GAI and LLM technologies.
SOURCE: HaystackID