Class Action Defense

By Gerald L. Maatman, and Jennifer A. Riley

Duane Morris Takeaway: In 2025, Artificial Intelligence – AI – continued to influence class action litigation on multiple fronts. First, we saw a growth of class action lawsuits targeting AI, including in the copyright area and employment space, as well as the securities fraud area with claims of “AI washing.” Second, we saw an increasing number of courts and lawyers err in their use of AI to generate documents filed on dockets across the country and encountered numerous examples of the ways in which AI is continuing to impact the efficiencies that underlie the litigation process.

DMCAR Editor Jerry Maatman discusses this trend in detail in the video below:

1. AI Provided Raw Material For Class Action Lawsuits

AI has been an accelerating force in class action litigation as a source of claims stemming from the development, use, and promotion of AI technologies. In 2025, some of those filed claims or ongoing claims included claims stemming from alleged copyright infringement, algorithmic bias or discrimination, and securities fraud.

On the copyright front, courts issued key decisions, including divergent decisions on whether using copyrighted works to train generative AI models constitutes “fair use” under the Copyright Act. In copyright cases, the plaintiffs typically allege that a developer of a generative AI tool violated copyright laws by using publicly available copyrighted works to train and inform the output of the AI tools. In Tremblay v. OpenAI, Inc., No. 23-CV-3223 (N.D. Cal. June 13, 2024), for instance, the plaintiffs alleged that OpenAI trained its algorithm by “copying massive amounts of text” to enable it to “emit convincingly naturalistic text outputs in response to user prompts.” The plaintiffs alleged these outputs included summaries that were so accurate that the algorithm must have collected and retained knowledge of the ingested copyrighted works in order to output similar textual content. The plaintiffs typically invoke the Copyright Act to allege that the defendant willfully made unauthorized copies of thousands of copyrighted works, generating damages up to $150,000 per copyrighted work for willful infringement, and, therefore, to seek billions in damages.

The $1.5 billion settlement reached in Bartz, et al. v. Anthropic is a landmark settlement and a prime example. In that suit, three authors filed a class action lawsuit against Anthropic claiming that Anthropic had downloaded millions of copyrighted books from “shadow libraries” like Library Genesis and Pirate Library Mirror to train its AI systems. In June 2025, Judge William H. Alsup of the Northern District of California denied Anthropic’s motion for summary judgment on the issue of fair use in a split-the-baby decision. The record showed that Anthropic downloaded more than seven million books from pirate sites but also bought and scanned millions more. The court held that Anthropic’s use of legally acquired books for AI training was protected fair use but that downloading and keeping pirated copies was not, noting that a developer that has obtained copies of books “from a pirate site has infringed already, full stop.” In August 2025, Judge Alsup granted the plaintiffs’ motion for class certification, sua sponte defining the class to include “all beneficial or legal copyright owners of the exclusive right to reproduce copies of any book” in the datasets that met his criteria. With tens of billions of dollars on the line, the parties promptly reached a settlement for $1.5 billion, the largest settlement of any class action in 2025.

Notably, shortly after Judge Alsup’s decision on summary judgment, Judge Vince Chhabria of the U.S. District Court for the Northern District of California reached a different conclusion in Kadrey, et al. v. Meta Platforms, Inc., No. 2023-CV-03417 (N.D. Cal. June 25, 2025). In that case, 13 authors, mostly famous fiction writers, sued Meta for downloading their books from online “shadow libraries” and using the books to train Meta’s generative AI models (specifically, its large language models, called Llama). The parties filed cross-motions for partial summary judgment regarding fair use. The court rejected the plaintiffs’ argument that “the fact that the AI developer downloaded the books from shadow libraries and did not start with an ‘authorized copy’ of each book gives them an automatic win.” The court held that, because Meta’s use of the works was highly transformative, to overcome a fair use defense, the plaintiffs needed to show that the AI model harmed the market for the plaintiffs’ works. Because the plaintiffs presented no meaningful evidence of market dilution, the court entered summary judgment for Meta on the fair use defense.

On the employment front, Mobley, et al. v. Workday, Inc., No. 23-CV-770 (N.D. Cal. May 16, 2025), continues to reign as one of the most watched and influential cases. In Mobley, the plaintiff, an African American male over the age of 40, who alleged that he suffers from anxiety and depression, brought suit against Workday claiming that its applicant screening tools discriminated against applicants on the basis of race, age, and disability. The plaintiff claimed that he applied for 80 to 100 jobs, and despite holding a bachelor’s degree in finance, among other qualifications, did not get a single job offer. The district court granted the defendant’s motion to dismiss on the ground that plaintiff failed to plead sufficient facts regarding the supposed liability of Workday as a software vendor for the hiring decisions of potential employers. In other words, the plaintiff failed to allege that Workday was “procuring” employees for its customers and merely claimed that he applied for jobs with a number of companies that all happened to use Workday.

On February 20, 2024, the plaintiff filed an amended complaint alleging that Workday was an agent of the employers that delegated authority to Workday to make hiring process decisions or, alternatively, that Workday was an employment agency or an indirect employer. Plaintiff claimed, among other things, that, in one instance, he applied for a position at 12:55 a.m. and his application was rejected less than an hour later. Judge Rita F. Lin granted in part and denied in part Workday’s motion to dismiss the amended complaint. The court reasoned, among other things that the relevant statutes prohibit discrimination “not just by employers but also by agents of those employers,” so an employer cannot “escape liability for discrimination by delegating [] traditional functions, like hiring, to a third party,” and an employer’s agent can be independently liable when the employer has delegated to the agent “functions [that] are traditionally exercised by the employer.” The court noted that, if it reasoned otherwise, and accepted Workday’s arguments, then companies could “escape liability for hiring decisions by saying that function has been handed to over to someone else (or here, artificial intelligence).”

The court opined that, given Workday’s allegedly “crucial role in deciding which applicants can get their ‘foot in the door’ for an interview, Workday’s tools are engaged in conduct that is at the heart of equal access to employment opportunities.” The court also denied Workday’s motion to dismiss the plaintiff’s disparate impact discrimination claims reasoning that “[t]he zero percent success rate at passing Workday’s initial screening” combined with the plaintiff’s allegations of bias in Workday’s training data and tools plausibly supported an inference that Workday’s algorithmic tools disproportionately rejected applicants based on factors other than qualifications, such as a candidate’s race, age, or disability. Thereafter, the court conditionally certified a collective action of all individuals aged 40 and over who applied for jobs using Workday’s platform and were rejected. In doing so, it authorized plaintiff to send notice of the lawsuit to applications nationwide. This litigation has been closely watched for its novel case theory based on artificial intelligence use in making personnel decisions and, given its success to date, is likely to prompt tag along and copycat litigation.

On the securities front, over the past three years, plaintiffs have filed dozens of lawsuits alleging that various defendants made false or misleading statements related to AI technology or related to AI as a driver of market revenue or demand, including claims that companies overstated their AI capabilities, effectiveness, or revenue generation in a practice known as “AI washing.” For instance, on April 17, 2025, the plaintiff Wayne County Employees’ Retirement System filed suit against AppLovin Corporation, No. 25-CV-03438 (N.D. Cal.), alleging that, among other things, the company falsely attributed its financial success to its enhanced AXON 2.0 digital ad platform and the use of “cutting edge” AI technologies to match advertisements to mobile games. In the complaint, the plaintiffs claim that the company’s revenue instead stemmed from manipulative ad practices, such as forced, silent app installations and that, upon release of short-seller reports disclosing the alleged practices, the company’s share price declined more than 12%.

Because investors have shown a willingness to pay a premium for shares of companies that appear positioned to capitalize on the effective use of AI, such statements have had the tendency to boost share prices. When projections fail to materialize, however, and share prices decline, plaintiffs are poised to take advantage.

In another example, plaintiffs filed a securities class action against Apple in the Northern District of California alleging that Apple made misleading and false statements regarding Siri’s generative AI features. The plaintiffs allege that Apple, at its annual Worldwide Developers Conference and on earnings calls, made claims that its AI solution called Apple Intelligence would create a more advanced and capable Siri. The plaintiffs allege that Apple continued to maintain that these features would arrive in early 2025 until March 2025 when it admitted that “[i]t’s going to take us longer than we thought.” The plaintiffs allege that, in the wake of these announcements, Apple’s share price dropped almost $47.

In sum, with AI continuing to flourish, the implications of its development, use, and advertisement are providing the raw material for creative plaintiffs’ class action lawyers. We should expect to see an upward trend of key decisions and new cases in 2026 and beyond as this burgeoning area of the law continues to expand.

2. AI Continued To Impact The Litigation Process

As legal professionals on both sides leverage AI to attempt to increase efficiency and gain a strategic advantage, examples of improper use abound. Rarely a day passes without a headline reporting attorney misconduct. To date, much of the AI misuse has centered on attorneys submitting or courts generating filings and legal briefs with fake citations. So-called “AI hallucinations” can take the form of citations to cases that do not exist or, even worse, the attribution of incorrect “hallucinated” holdings or quotations to existing opinions.

Bar associations have compiled dozens if not hundreds of instances of attorneys misusing generative AI in complaints, legal memoranda, expert reports, and appellate briefs. Perhaps more disturbing, these examples are joined by at least two instances of courts withdrawing decisions due to the incorporation of AI-generated contend.

Such conduct has led to severe sanctions, including fines and suspensions for violation of ethical duties, as well as (presumably) terminations. To date, claims of overbilling for such AI-generated worked product have not been made public, and lawyers continue to reiterate and train that AI is a tool and not a substitute for the application of legal analysis and judgment.

At the same time, AI is becoming an asset in the hands of more cautious connoisseurs who are taking advantage of its efficiencies for projects involving data analytics, document reviews, and form generation. Its use has become transformative in the settlement administration process where it has exposed vulnerabilities in the claims administration process by, for example, generating thousands of entries that dilute legitimate claims, thereby reducing legitimate recoveries.

Similar to classrooms where teachers use AI to detect AI, recipients are responding with their own AI-based tools to detect irregularities.

As the technology continues to evolve, it continues to impact that class action space in particular, which is particularly susceptible to mass-generated claims, demand letters, and form complaints. As a result, we are likely seeing the tip of the iceberg in terms of AI’s influence on the class action space.

By Gerald L. Maatman, Jr. and Jennifer A. Riley

Duane Morris Takeaway: In 2025, case law continued to develop in fragmented ways among the federal circuits on issues material to plaintiffs’ ability to maintain and certify class actions, enhancing the likelihood of and incentive for forum shopping. In terms of standards governing conditional certification of FLSA, EPA, and ADEA matters, 2025 saw the crystallization of four distinct standards, ranging in the burdens applicable to plaintiffs, as well as in the review and consideration of the evidence presented. A second chasm relates to courts’ approaches uninjured class members, or the notion that each member of a putative class as defined might not have experienced a concrete injury sufficient to provide such individual standing to pursue a claim. A third chasm reflects courts’ divergent views relative to personal jurisdiction and whether a court that cannot exercise general personal jurisdiction must have a basis for specific personal jurisdiction as to each putative class member.

DMCAR co-editor Jennifer Riley explains this trend in detail in the video below:

These fractures have made forum selection more consequential than ever. Plaintiffs are increasingly skewing their filings toward federal circuits where they anticipate a greater likelihood of a favorable outcome, including toward jurisdictions where judges are taking a more lenient approach to certification or a more permissive view on issues like standing and jurisdiction. To date, efforts to persuade the U.S. Supreme Court to take up cases that would resolve these splits have failed, so we expect they will continue to drive uncertainty in class-related litigation through 2026.

1. Courts Disagree Over The Standards For Conditional Certification Of Collective Actions

The standards for conditional certification under the FLSA, EPA, and ADEA, continue to diverge such that district and appellate courts are applying any of at least four distinct approaches. These statutes provide little guidance as to the process they intended to incorporate for so-called conditional certification. In 29 U.S.C. § 216(b), the FLSA provides that “[a]n action . . . may be maintained against any employer (including a public agency) . . . by any one or more employees for and in behalf of himself or themselves and other employees similarly-situated. No employee shall be a party plaintiff to any such action unless he gives his consent in writing to become a party and such consent is filed in the court in which such action is brought.” 29 U.S.C. § 216(b). Courts have interpreted such language to authorize a process by which courts grant “conditional certification” of a collective action, authorize notice to persons who fall within the defined group, and permit those persons to “opt-in” by returned their consent forms. Courts, however, have disagreed over the standards plaintiffs must satisfy to initiate this process. In other words, how and when should a court determine if such persons are “similarly-situated”? 

To date, federal courts that have addressed these issues have developed or adopted one of four primary schemes.

First, for many years, court accepted the familiar and lenient two-step standard set forth in Lusardi v. Xerox Corp., 118 F.R.D. 351 (D.N.J. 1987), and the U.S. Court of Appeals for the Second Circuit expressly adopted this standard in Scott v. Chipotle Mexican Grill, Inc., 954 F.3d 502, 515 (2d Cir. 2020), while the First, Third, Tenth, and Eleventh Circuits had done the same by “acquiescence” without express adoption. See Kwoka v. Enterprise Rent-A-Car Company of Boston, LLC, 141 F.4th 10, 22 (1st. Cir. 2025); Zavala v. Wal Mart Stores Inc., 691 F.3d 527, 534 (3d Cir. 2012); Thiessen v. General Electric Capital Corp., 267 F.3d 1095, 1105 (10th Cir. 2001); Hipp v. Liberty National Life Insurance Co., 252 F.3d 1208, 1219 (11th Cir. 2001)

Under Lusardi, a court considers at “step one” whether a plaintiff has made a “modest factual showing” based on his or her evidence, which often comprises one or more declarations, and may or may not even look at competing evidence submitted by the employer. If the court determines that a plaintiff has satisfied his or her “lenient” burden, the court authorizes notice. At the close of discovery, the employer then can move to decertify the conditionally certified collective action, and the court will consider based on the evidence whether the plaintiff has demonstrated that the persons who joined the action are similarly situated.

The Ninth Circuit clarified in Campbell v. City of Los Angeles, 903 F.3d 1090, 1114 (9th Cir. 2018), that the plaintiff must show he or she is similarly situated with respect to “some material aspect” of his or her claim and not merely in some way that is irrelevant to the claims asserted.

Second, in the first example of a court revisiting and examining the text of the FLSA, the Fifth Circuit prompted the ensuing split with its decision in Swales v. KLLM Transportation Services, LLC, 985 F.3d 430, 443 (5th Cir. 2021). In that decision, the Fifth Circuit rejected Lusardi’s two-step approach outright and directed district courts to “rigorously enforce” the FLSA’s similarity requirement at the outset of the litigation in a one-step approach. “[T]he district court needs to consider all of the available evidence” at the time the motion is filed and decide whether the plaintiff in fact has “met [his or her] burden of establishing similarity.” Id. at 442-43.

Third, in the wake of Swales the Sixth Circuit likewise revisited the standard in Clark v. A&L Homecare & Training Center, LLC, 68 F.4th 1003 (6th Cir. 2023). The Sixth Circuit rejected Lusardi, but also declined to adopt Swales. Instead, the court likened the standard to one comparable to the standard for obtaining a preliminary injunction. An employee must show a “strong likelihood” that others are similarly situated to the employee before the district court may authorize the plaintiff to send notice of the action. The Sixth Circuit left open the standard by which the court should consider a potential motion for decertification down the line. Id. at 1011.

Fourth, most recently, the Seventh Circuit addressed the same issue in Richards, et al. v. Eli Lilly & Co., 149 F.4th 901 (7th Cir. 2025). The Seventh Circuit rejected the Lusardi framework but declined to go as far as Swales or Clark. Instead, the Seventh Circuit ruled that “a plaintiff must first make a threshold showing that there is a material factual dispute as to whether the proposed collective is similarly situated” to secure a ruling authorizing notice, and an employer “must be permitted to submit rebuttal evidence” for the court to consider. Id. at 913. The court declined to set any bright line rule as to whether a court should decide the similarly situated question in a one or two step approach, noting that the analysis is not an “all-or-nothing determination.” Id. at 913-914.

The U.S. Courts of Appeal for the District of Columbia, Fourth, and Eighth Circuits have not yet opined on the proper method, leaving district courts to exercise their discretion. These divergent standards have influenced forum selection, as plaintiffs significantly have decreased the number of collective actions they pursue in the Fifth and Sixth Circuits in particular, in favor of filing in forums that apply more lenient standards.

2. Courts Continue To Disagree Over Standing And Personal Jurisdiction

Courts continue to disagree regarding the impact and treatment of uninjured class members, a key issue that remains unresolved. It is axiomatic that individuals who did not suffer injury as the result of the defendant’s conduct cannot maintain claims, and courts do not have the power to award them relief. As the U.S. Supreme Court reiterated in its seminal 2020 decision in TransUnion, “Article III does not give federal courts the power to order relief to any uninjured plaintiff, class action or not.” TransUnion LLC v. Ramirez, 141 S.Ct. 2190, 2208 (quoting Tyson Foods v. Bouaphakeo, 577 U.S. 442, 466 (2016) (Roberts, C.J., concurring)). In this respect, the “plaintiffs must maintain their personal interest in the dispute at all stages of the litigation . . . And standing is not dispensed in gross; rather, plaintiffs must demonstrate standing for each claim that they press and for each form of relief that they seek.” Id.

Despite this admonition, courts continue to grapple with the application of these concepts in the class certification context and, in particular, they disagree over whether to certify a class, a plaintiff must demonstrate that every putative class member has standing, or, stated differently, must demonstrate that the class excludes those individuals who did not suffer harm. In TransUnion, the Supreme Court expressly left open the question of “whether every class member must demonstrate standing before a court certifies a class.” Id. at n.4. Such a requirement has significant consequences for the class action landscape.

As a result, in January 2025, the U.S. Supreme Court granted a petition for certiorari in Laboratory Corporation Of American Holdings v. Davis, 145 S.Ct. 1608 (2025). In Davis, plaintiffs filed suit on behalf of a putative class of legally blind patients alleging that Lab Corp. violated the ADA by failing to make its self-service check-in kiosks accessible. In May 2022, the district court certified a broad class of that included all legally blind individuals were denied full and equal enjoyment of its goods and services due to “LabCorp’s failure to make its e-check-in kiosks accessible,” emphasizing that individualized damages questions do not defeat the predominance requirement. Lab Corp. sought interlocutory appeal, arguing that plaintiffs’ class definition swept in uninjured individuals who would not have used kiosks anyway. The Ninth Circuit granted the petition and affirmed. Applying Ninth Circuit precedent, the appellate court reasoned that Rule 23 permits certification of a class even when the class “‘potentially includes more than a de minimis number of uninjured class members.’” The U.S. Supreme Court granted certiorari in January 2025. Following briefing and oral argument, the U.S. Supreme Court declined to resolve the issue and dismissed the writ as improvidently granted. Justice Kavanaugh authored a dissent from such decision noting that, if given the opportunity, he would hold that “[f]ederal courts may not certify a damages class under Rule 23 when, as here, the proposed class includes both injured and uninjured class members.”

Without guidance from the U.S. Supreme Court, lower federal courts have continued to reach varying decisions on the issue. For instance, on July 17, 2025, the Fifth Circuit issued its decision in Wilson v. Centene Management Co., 144 F.4th 780 (5thCir. 2025). The plaintiffs in this case asserted breach of contract claims against the defendant insurance companies, alleging that it issued inaccurate provider lists and thereby caused the plaintiffs to pay artificially inflated premiums for access to providers who were not available. The district court denied class certification finding that the plaintiffs lacked standing. On appeal, the Fifth Circuit held that, at the class certification state, a plaintiff need only demonstrate his or her own standing, and the district court erred in its determination of the plaintiff’s standing, which it reached through a merits-based evaluation of the plaintiff’s expert.

The Seventh Circuit addressed the question in Arandell Corp. v. Xcel Energy Inc., 149 F.4th 883 (7th Cir. 2025). In that case, the plaintiffs brought a putative state-wide class action alleging that defendants engaged in a price-fixing conspiracy to manipulate natural gas prices. The Seventh Circuit noted that, to the extent defendants suggested that, before class certification, the plaintiffs must show all class members suffered some injury, “that is not correct.” It reiterated its prior holding that that “a class should not be certified if it is apparent that a great many persons who have suffered no injury at the hands of the defendant,” it clarified that “[t]here is no precise measure for ‘a great many.’  Such determinations are a matter of degree and will turn on the facts as they appear from case to case.”

In contrast, the Fourth Circuit took a different approach in Freeman v. Progressive Direct Insurance Co., 149 F.4th 461 (4th Cir. 2025). After an automobile collision, the plaintiff’s insurer provided her a payment based on the “actual cash value” of her car that it determined using a “projected sold adjustment.” Although the plaintiff accepted the payment, and did not contest the valuation, she filed suit for breach of contract. Although the district court certified a class, the Fourth Circuit reversed. The Fourth Circuit explained that, to succeed on her claim, the plaintiff needed to show that her insurer paid her less than the actual cash value of her vehicle and, likewise, that her insurer paid members of the class less than the actual cash value of their vehicles, regardless of whether the “projected sold adjustment” was used in determining that value. Yet, the class was defined to include anyone who was paid “compensation for the total loss of a covered vehicle, where . . . the actual cash value was decreased based upon Projected Sold Adjustments.” Thus, the class was defined to include insureds who accepted the insurer’s offer of payment, insureds who negotiated a higher payment, and insureds who invoked the appraisal process in the policy, simply because in each circumstance the insurer made its calculation using the Projected Sold Adjustment. “Yet, none of those could claim injury because each agreed to resolution of the loss. . . This characteristic of the certified class alone justifies reversal of the class certification order.”  

Similarly, courts have continued to disagree regarding the scope of a court’s personal jurisdiction over the defendant in the class action context. In short, the U.S. Supreme Court decided Bristol Myers Squibb v. Superior Court, 137 S. Ct. 1773 (2017), in 2017 and ruled that a court must have a basis for exercising personal jurisdiction over a defendant for each claim it adjudicates. In that case, which involved a mass tort action, the U.S. Supreme Court concluded that the existence of similar claims asserted by plaintiffs who purchased a drug in California did not provide a court with personal jurisdiction over the defendant for purposes of adjudicating claims asserted by plaintiffs who purchased the same drug outside of California.

Again, despite this clear ruling, courts have continued to grapple with the application of these concepts in the class certification context and, in particular, they disagree over whether, to certify a class that includes nationwide class members, a plaintiff must demonstrate that the court can exercise personal jurisdiction over the defendant for purposes of resolving each of their claims or, stated differently, must demonstrate that each claim arises from or relates to a foreign defendant’s contacts with the forum state.

On July 1, 2025, the Ninth Circuit became the latest to address this issue. In Harrington, et al. v. Cracker Barrel Old Country Store, 142 F.4th 678 (9th Cir. 2025), the Ninth Circuit ruled that the U.S. Supreme Court’s decision in Bristol-Meyers applies to collective actions brought under the FLSA. The Ninth Circuit held that, when a plaintiff relies on specific personal jurisdiction as the basis for personal jurisdiction over the defendant in an FLSA collective action, district courts must assess whether they can exercise specific personal jurisdiction over the defendant on a claim-by-claim basis. This means that the claim of every opt-in plaintiff must arise out of or relate to the defendant’s activities in the forum state, and opt-in plaintiffs with no connection to the forum cannot rely on the connections of the named plaintiffs to establish personal jurisdiction.

In sum, courts continue to disagree as to their power in the class action context and the extent to which a procedural rule like Rule 23 can alter otherwise fundamental concepts of subject matter and personal jurisdiction for putative class members. Given the implications of such rules, we can anticipate that such questions will continue to influence forum selection for plaintiffs and continue to fuel uncertainty for defendants through 2026.