The AI Layoff Lawsuit That Should Worry Every HR Team

Twenty-six people did not expect to learn why they lost their jobs from a court filing. In May 2026, Meta cut roughly 8,000 positions, about 10% of its workforce, in a reduction in force the company described as routine. By July, a federal lawsuit was arguing something else: that the algorithm doing the ranking could not tell the difference between an employee falling behind and an employee on approved medical leave, and that Meta never built in a way to correct for it.

What the Lawsuit Says Meta's AI Missed

The plaintiffs, all current or former Meta employees who had taken or requested protected medical or family leave, filed suit in federal court in Oakland, alleging Meta relied on a "constellation of internal artificial intelligence systems" to score, rank, and select employees for the May reduction in force, including tools that monitored keystrokes and computer activity, tracked AI token usage, and generated algorithmically assisted performance rankings. The complaint's core claim is structural rather than incidental: the scoring systems, by design, could not account for an employee whose output dropped because they were on leave or working under a disability accommodation, so protected time was recorded as underperformance and fed directly into who got cut. Among the plaintiffs were eight women on maternity or pregnancy leave, four men on parental leave, and a woman on bereavement leave. One plaintiff, a scientist on approved pregnancy leave, learned she had been selected for termination two days before giving birth.

The complaint brings claims under the Family and Medical Leave Act, the Americans with Disabilities Act, the Pregnancy Discrimination Act, and the Pregnant Workers Fairness Act. Meta has denied the allegations, telling reporters that "workforce management decisions were and are made by people, not AI."

Two Rulings, No Relief, and a Case That Isn't Over

Meta hasn't lost in court, and that is precisely why HR teams elsewhere should be paying attention. On July 17, 2026, Judge William Orrick declined to issue a temporary restraining order pausing the layoffs, finding the plaintiffs hadn't shown the kind of irreparable harm emergency relief requires. Lost health insurance, protected leave time, and unvested equity, he wrote, "can be remedied through damages, back pay or other comparable relief in the arbitration process." He carved out one exception: four plaintiffs on temporary work visas faced possible removal from the country within 60 days without new employment, a harm he called more than speculative, and he ordered Meta to explain why those particular employees were selected.

The plaintiffs' broader request for a preliminary injunction fared no better. In an August ruling, Orrick said "the record at the moment does not persuade me of the merits," while acknowledging the plaintiffs' evidence "raised some potential questions about Meta's categorical denial of any impact of AI in the termination process." Rather than resolve those questions himself, he pointed the case toward arbitration, where the plaintiffs can still pursue an independent audit of how Meta's layoff decisions were actually made. Twice denied emergency relief is not the same as twice cleared. The mechanics of Meta's AI systems still haven't been tested in full.

Meta Isn't the Outlier, It's the Preview

What makes the case worth reading outside Menlo Park is how ordinary the underlying practice turns out to be. A July 2026 survey of 1,000 U.S. managers found 59% already use AI to help decide who gets laid off, and 24% do so often or all the time. Among the factors those tools weigh: 80% consider performance or productivity, which is exactly the metric the Meta plaintiffs say penalized their leave time, and 31% factor in sick days or medical leave directly. Forty-three percent of managers let AI make the layoff call with no human review at least occasionally, 17% do so often or always, and 58% could not confirm the tool they were using had ever been tested for bias. Thirty-eight percent said they'd received no training on using AI ethically in HR decisions at all.

The legal theory behind the Meta case, and behind the separate Mobley v. Workday litigation moving through the same courts, is that buying or building the AI doesn't buy immunity from it. Courts have increasingly treated screening and scoring algorithms as an extension of the employer making the decision, not a neutral third party, which means the standard defense, that the tool did it, not us, is losing ground. Colorado's SB 24-205 goes further, placing affirmative obligations on employers using "high-risk" AI systems to actively prevent algorithmic discrimination rather than simply respond to it after the fact. And AI-driven processes leave a more complete paper trail than a manager's judgment call ever did: every score, every weighting, every excluded factor is logged and reconstructable, which cuts against employers in discovery exactly when they need ambiguity most.

• Audit what your layoff or performance-scoring tools actually weigh. With 31% of managers admitting sick days or medical leave factor into AI layoff scores, the operative question isn't whether AI touches your RIF process, it's whether protected time is quietly counted against someone.
• Never let AI make an unsupervised termination call. Forty-three percent of managers already do, and the Meta case shows exactly what a court asks to see once a human never checked the ranking before it went out.
• Build a leave-neutral review into every reduction in force, before it goes out the door. A documented check for whether protected leave or disability status shaped a selection list is cheap insurance against the claim Meta is now defending in court.
• Treat a vendor's AI as your liability, not theirs. The legal theory unfolding in both the Meta and Workday cases treats scoring tools as an extension of the employer's own decision-making, whether you built the model or licensed it.
• Know the metrics well enough to defend them under discovery. AI-driven decisions generate a more complete record than a manager's judgment call ever did, which means HR needs to be able to explain every input before outside counsel has to.
• Apply the same standard across borders, not just U.S. headcount. Multinational employers restructuring through M&A or global workforce changes face this exposure in every jurisdiction where the AI-assisted decision lands.

Meta has not been found liable for anything. What it has done, twice, is fail to make the questions about its AI go away. For every HR team running layoff scores through a tool nobody has stress-tested for bias, that is the part of this story worth sitting with.