Twenty-six former Meta employees have sued Mark Zuckerberg’s company. They claim that during the spring layoffs, the company had computer systems compile a list of people to be dismissed and that this approach disproportionately affected employees with disabilities and those who took medical or parental leave. It appears to be the first case of its kind in which someone has challenged the use of artificial intelligence in the layoff process at a major American company.
At the beginning of the year, Meta announced that it would reduce its workforce by one-tenth, or roughly eight thousand people. The wave of layoffs began in May, with further cuts expected later. The twenty-six plaintiffs are among those laid off. In May, they learned that their positions would be eliminated as of July 22.
Decisions by systems instead of managers
According to the seventy-page complaint, Meta did not compile the list based on the judgment of supervisors who were familiar with their employees’ work. Instead, it allegedly relied on what the plaintiffs call a constellation of internal artificial intelligence systems. These systems were supposed to score and rank employees and select those to be placed on the departure list.
What did this setup include? An internal language assistant called Metamate. A so-called second brain, an agent that employees themselves trained by providing it with their communications and documents so that it could learn to imitate their work. An activity-monitoring system that continuously collected data on keystrokes, screen content, mouse movement, browser history, emails, and messages on company devices. It also included an overview of the consumption of so-called tokens, a measure of how extensively each person used the company’s artificial intelligence tools, including a leaderboard comparing individuals with one another. And finally, an algorithmic calibration tool for performance reviews that, according to the lawsuit, largely replaced the previous manager-led evaluation process.
A number of indicators were included in the score. Annual performance reviews for the previous twelve months, the number of code changes uploaded, the level of artificial intelligence tool usage, output volume, a supervisor’s recommendation, and indicators of alignment with the company’s strategic plans.
Data that cannot be collected while on medical leave
This is precisely the main objection. The plaintiffs argue that, by their very nature, these indicators cannot be accumulated by someone who is on medical leave or vacation, or whose output is reduced by a disability.
Someone recovering from surgery is not typing on a keyboard. Someone on parental leave is not consuming company tokens. An employee with an approved workplace accommodation due to a disability will not produce the same volume of trackable activity as a colleague working at full capacity. During such an absence, both keystroke frequency and token consumption decline.
According to the complaint, Meta did not account for time spent on leave in any way, did not exclude it from the measured indicators, and did not ensure that people on leave or with workplace accommodations received an individual human assessment, as the plaintiffs say the law requires. The company thus allegedly disproportionately selected those who had taken leave or requested disability-related accommodations.
Four laws and the question of bias
The plaintiffs accuse Meta of violating several regulations. They cite the Family and Medical Leave Act, the Americans with Disabilities Act, and the Pregnancy Discrimination Act. They also add another allegation. Meta allegedly failed to test its systems for bias, as required by recently enacted laws in California and New York.
The case also involves one troubling detail from the past. The monitoring program in question was suspended in June after a security failure exposed employees’ private conversations, transcripts, and performance data to the entire company. For the plaintiffs, this is evidence that the behavioral monitoring systems did in fact capture and store sensitive personal performance data and that Meta’s management of this data was, at least at the time, inadequate.
Meta: People, not machines, made the decisions
The company denies wrongdoing. A Meta spokesperson said that the allegations are not supported by the facts and that workforce management and organizational change decisions were and continue to be made by people, not artificial intelligence. Meta attorney Erin Connell also challenged the claim of imminent harm in court. According to her, employees are not losing their health insurance altogether; they are only losing the portion subsidized by the employer.
The plaintiffs’ attorneys saw it differently. Barbara Cowan reminded the judge that along with their jobs and wages, people would also lose valuable stock options and insurance, jeopardizing their medical treatment or pregnancies, or complicating the management of other health issues. According to her, childbirth and the first weeks with a child cannot be undone.
Judge declined the request but left the door open
The plaintiffs sought a temporary injunction that would prevent Meta from completing the layoffs until their claims were decided in private arbitration. Federal Judge William Orrick in Oakland ruled in writing that he would not prevent the company from proceeding with the layoffs beginning July 22. According to him, the employees failed to demonstrate that losing their jobs constituted the so-called irreparable harm required by law for such an order to be issued.
At the same time, Orrick indicated that he might reconsider his position once he had more information about how the layoffs were carried out. If the plaintiffs demonstrate during arbitration that the decisions were indeed driven by artificial intelligence, the judge will revisit the matter. In a joint statement, the plaintiffs’ attorneys said that although Orrick denied their request, he also acknowledged that the dispute raises serious questions about Meta’s conduct.
Meta had faced a similar allegation shortly before, when a former employee accused it of disproportionately affecting older workers in its February 2025 round of layoffs. One of the engineers in the current lawsuit also claims that he knew of employees who were laid off while on paternity leave. Employee activity monitoring is by no means solely a Meta problem. According to a 2025 survey, more than seventy percent of employers in the United States said they used some form of online tool to monitor their employees’ activities.
Sources: latimes.com, reuters.com, theguardian.com and wsj.com
