The Commonality Crucible: Workday's AI Certification Fight Defines the Next Class Action Battleground

The Algorithmic Black Box Meets Rule 23

The most consequential development in this week’s civil litigation landscape isn’t an appellate reversal or a Supreme Court grant—it’s a district court certification fight that could determine whether algorithmic discrimination claims can survive as class actions in the federal courts. On November 11, Workday told a federal judge that millions of job applicants cannot proceed as a group in their proposed class action alleging that the company’s AI screening tools systematically discriminate based on race, sex, age, and disability.

Workday’s argument is deceptively simple and, if accepted, could functionally immunize AI vendors from class-wide employment liability: the named plaintiffs lack common evidence because the company’s algorithms are not monolithic. Different employers configure the tools differently; different applicants interact with different versions of the software; and the AI’s decision-making process varies too substantially across millions of hiring decisions to satisfy the commonality and predominance requirements of Fed. R. Civ. P. 23.

This is not merely a procedural skirmish. It is a direct assault on the viability of class actions in the age of machine learning, and it arrives at a moment when plaintiffs’ firms are racing to hold AI developers accountable under Title VII, the ADEA, and the ADA. If Workday prevails, the message to the plaintiffs’ bar will be clear: algorithmic discrimination is too individualized, too opaque, and too variable to certify.

The Falcon Problem and the "Black Box" Defense

Workday’s position leans heavily on the Supreme Court’s decision in Wal-Mart Stores, Inc. v. Dukes, 564 U.S. 338 (2011), which rejected certification of a nationwide class of female employees because the plaintiffs could not identify a common, company-wide policy of discrimination. The Court held that discretionary decisions by local managers, standing alone, could not supply the "glue" necessary to hold a class together under Rule 23(a)(2).

Workday is effectively arguing that its AI is the digital equivalent of thousands of individual store managers—except worse for plaintiffs, because the algorithmic "decisions" are not even uniform within a single employer’s hiring process. The company contends that the plaintiffs cannot point to a common mode of exercising discretion (or, in this case, a common algorithmic output) that unites the class.

This creates a catch-22 that should concern every employment lawyer representing either side. To prove commonality, plaintiffs need discovery into how the algorithm functions. But Workday—and the broader tech industry—will resist that discovery on trade-secret grounds, citing the need to protect proprietary models. Without transparency into the training data and decision logic, plaintiffs may be unable to satisfy Comcast Corp. v. Behrend, 569 U.S. 27 (2013), which requires that damages be measurable on a class-wide basis through a common methodology.

The practical consequence: If certification is denied because the AI is too complex to evaluate as a unified system, defendants will have discovered the perfect antitrust and discrimination shield—complexity itself becomes a defense to class treatment.

Why This Matters Beyond Employment Law

The Workday certification fight is not an isolated event. It is the leading edge of a broader judicial skepticism toward algorithmic class actions that is playing out across multiple circuits. Compare the Third Circuit’s July revival of an antitrust class action against Atlantic City casinos—where plaintiffs alleged that an AI revenue platform coordinated room prices—with the Ninth Circuit’s earlier rejection of similar claims in a Nevada case. The circuit split suggests that appellate courts are struggling with whether algorithmic coordination can constitute a "common" antitrust conspiracy under Bell Atlantic Corp. v. Twombly, 550 U.S. 544 (2007), or whether the AI’s independent decision-making breaks the causal chain.

Similarly, the November 9 decision allowing deceptive advertising claims to proceed against Anthropic—while dismissing others—demonstrates that judges are willing to parse which AI-related claims are susceptible to common proof and which require individualized inquiries. The trend is unmistakable: courts are becoming more granular, and more hostile, in their Rule 23 analysis of technology-driven claims.

The Strategic Imperative for Practitioners

For plaintiffs’ attorneys, the Workday filing is a warning shot. The era of filing broad, nationwide classes based on the mere existence of an AI tool is over. To survive certification, lawyers must:

1. Demand algorithmic audits early. Use Fed. R. Civ. P. 26(b)(1) to seek not just outcomes data, but model architecture, training datasets, and validation studies. If the defendant claims trade secret protection, move for protective orders that allow expert review under confidentiality restrictions.

2. Narrow the class definition aggressively. Instead of "all applicants rejected by Workday’s AI," consider "all Black applicants over 40 rejected by Client X’s implementation of Workday’s resume-screening module between 2023 and 2025." Smaller, more homogeneous classes have a better chance of satisfying Amchem Products, Inc. v. Windsor, 521 U.S. 591 (1997), which requires that common questions predominate over individual ones.

3. Focus on the "choice architecture," not just the output. Even if the AI’s decisions vary, the vendor’s design choices—what data to include, what proxies to use, what thresholds to set—may constitute common, class-wide conduct that satisfies Rule 23(b)(3).

For defense counsel, Workday’s strategy offers a roadmap, but also a risk. If courts begin accepting the "black box" defense at the certification stage, expect plaintiffs to respond by seeking broader injunctive relief under Rule 23(b)(2)—which does not require predominance—or by filing individual actions that could expose companies to greater aggregate liability and inconsistent judgments.

The Bottom Line

The Workday certification fight will likely be resolved in early 2026, but its impact will be felt immediately. Employment lawyers should advise clients to audit their AI vendors now, not after the class action is filed. If your client uses third-party screening tools, ensure the contract includes indemnification for discrimination claims and, more importantly, provisions requiring the vendor to cooperate in certification discovery.

The plaintiffs’ bar is watching closely. If Workday succeeds in decertifying millions of job applicants, we will see a fragmentation of AI litigation into thousands of individual arbitrations or single-plaintiff suits—precisely the outcome that Rule 23 was designed to prevent. The question for the federal courts is whether they will allow algorithmic complexity to become a de facto immunity, or whether they will adapt Rule 23 to the realities of machine decision-making.