When a group of former Meta employees began comparing notes after their departures, they concluded that the decisions had not been made by their managers. Now they are suing to prove it. A class action filed by former workers accuses Meta of using AI systems to evaluate employee performance and to select people for layoffs, alleging that the process discriminated against certain groups in ways that human managers, and the law, would not allow.
The lawsuit lands at the intersection of two trends that have defined Meta’s recent history: aggressive cost-cutting and heavy reliance on automation. Meta has eliminated tens of thousands of jobs since 2022, and in January of last year Chief Executive Mark Zuckerberg told employees the company would cut about 5 percent of its workforce for performance, saying it intended to raise the bar on retention. The company has used automated performance-review tools across those exercises, and the plaintiffs argue that the algorithms, not their managers, effectively made the calls about who stayed and who went.
The core difficulty, as Reuters noted in its analysis of the case, is one of proof. Under U.S. employment law, discrimination claims require showing that a protected characteristic played a role in an adverse decision, and in an algorithmic system that role is hard to demonstrate. The plaintiffs must show not just that Meta used AI, which the company does not deny, but that the AI produced outcomes that disadvantaged a protected group, and that the company cannot justify those outcomes by reference to legitimate performance differences.
The black-box problem cuts both ways. If employers cannot explain what an algorithm did, they also cannot easily prove that it did nothing wrong, and courts are still working out what standard of explanation to demand. Meta is expected to argue that its AI tools were aids to human decision-making, that managers reviewed the outputs and made the final calls, and that the company’s processes complied with its legal obligations. The plaintiffs will counter that when decisions are made at scale, with hundreds of thousands of employees and algorithmic ranking, the human review is procedural rather than substantive.
The case is one of a wave testing how employment law applies to AI. Companies across the economy use algorithms to screen résumés, score performance and predict attrition, and regulators have been circling the practice for years. The Equal Employment Opportunity Commission has warned that AI tools can violate anti-discrimination law if they produce disparate impact, and New York City has enacted one of the first laws requiring audits of automated employment decision tools. A case that reaches trial on the question of who made the decision, the algorithm or the manager, would give courts a chance to set rules that apply far beyond Meta.
The discovery process, if the case reaches it, could be the most consequential part of the litigation. Meta’s performance systems process data on hundreds of thousands of employees, and the plaintiffs will seek access to the models, the training data and the audit trails behind the layoff decisions. Companies have resisted such disclosure in other AI cases, arguing that their systems are trade secrets, and courts have had to balance that claim against plaintiffs’ need for evidence. How that balance is struck in this case will determine how much visibility workers, and their lawyers, get into the algorithms that affect their careers.
For Meta, the stakes are both legal and reputational. The company has defended its performance system as a fair way to manage a workforce at its scale, and its lawyers are likely to emphasize the layers of human oversight in the process. But the lawsuit touches a sensitive point: Meta’s culture of data-driven management, applied to the people who work there, has produced complaints for years, and the class action converts those complaints into a formal test of whether the company’s methods are legal.
Experts in employment law said the case faces long odds at the early stages, because class certification requires showing that common questions dominate individual ones, and performance decisions are notoriously individual. But even a case that fails on certification can change behavior, they added, because it puts employers on notice that algorithmic decision-making will be scrutinized, and because discovery can force companies to reveal how their systems actually work.
The broader question the lawsuit raises is how much authority an algorithm should hold over a career. Meta’s use of AI in performance management is not unusual, but it is unusually visible, and the company’s size makes it a natural test case for the rules that will govern AI in the workplace. If the case proceeds, the discovery process alone could force Meta to open the black box, showing how its systems ranked employees and how much weight the rankings carried in the decisions that followed.
For the former employees who filed the suit, the goal is compensation and, in their telling, accountability. For the industry watching, the case is a marker: the first generation of AI-driven layoffs is now old enough to produce lawsuits, and the second generation will be shaped by how these cases come out.
