AI Decided Who Got Fired — And Nobody Knows Who to Blame

On May 20, 2026, a Meta scientist received an email at 4 a.m. notifying her that her position had been eliminated. She was on approved prenatal leave at the time. The next day, her water broke. The day after that, she gave birth.

This is not a hypothetical. It is documented in a 71-page federal lawsuit filed by 26 Meta employees on July 13, 2026, in federal court in Oakland, California.

The plaintiffs — engineers, scientists, designers, researchers, managers, and directors across six states and Washington, D.C. — allege that Meta used a constellation of internal AI systems to select the roughly 8,000 employees it laid off in May, or about 10 percent of its workforce.

Their commonality: all 26 had taken or requested protected leave — pregnancy, parental, medical, or disability accommodation — within the previous 24 months.

The ‘Constellation’ of Systems

According to the complaint, Meta used a suite of AI-powered tools to score and rank employees. The plaintiffs describe it as a “constellation”:

Metamate: Meta’s internal large language model assistant.

Second Brain: A personal agent employees were required to train, ingesting their own communications and files to learn to replicate their output.

Employee activity monitoring: Continuous collection of keystrokes, screen content, mouse activity, browser history, emails, and messages on company devices.

AI token usage dashboards: Employee-level tracking of AI tool usage, ranked against colleagues.

Algorithmic performance calibration tools: Largely replacing manager-led performance reviews.

The inputs included rolling 12-month performance ratings, code commits, AI tool usage, output volume, manager endorsements, and roadmap alignment signals.

The problem is structural: an employee on 12 weeks of parental or medical leave cannot accumulate these metrics. According to the lawsuit, Meta did not exclude protected leave periods from the metrics, did not exclude leave-takers from the candidate pool, and did not pause the system for individual circumstances.

The system saw “leave” as “low output.” And low output meant selection.

The Human Cost

The complaint documents individual cases:

Doe 17, the scientist on prenatal leave, was in an organization where three people were selected for layoffs — two of whom were on pregnancy- or birth-related leave.

Doe 8, a designer whose maternity leave was approved through September, gave birth in March. On May 20, with a weeks-old infant at home, she received her termination notice.

One employee disclosed a “serious health condition and disability” approved by Meta’s own provider, but was allegedly “discouraged and deterred” from taking leave by a manager who warned it would result in layoff selection.

Of the plaintiffs, eight are women who took maternity or pregnancy-related leave, four are men who took parental leave, and one is a woman who took leave to care for a family member and later bereavement leave.

The lawsuit alleges violations of the Family and Medical Leave Act, the Americans with Disabilities Act, the Pregnancy Discrimination Act, and the Pregnant Workers Fairness Act.

Meta’s Defense

Meta has denied the allegations. “The claims lack merit and are not based on facts,” a company spokesperson said. “Workforce management and organizational decisions were and are made by people, not AI.”

The plaintiffs’ response is pointed: if managers were making the decisions, why were so many surprised by who was selected? Several plaintiffs reported that their direct managers were unaware of their selection — some had even been told the day before that they did not know who would be affected.

One analysis put it this way: Meta’s denial “isn’t as reassuring as it sounds because we don’t know for sure if a human wholly made the decision.”

The Legal Framework Gap

The lawsuit invokes a concept called “disparate impact liability” — a principle codified in Title VII of the 1964 Civil Rights Act. It holds that facially neutral policies or practices can be discriminatory if they disproportionately burden a protected class and are not necessary for the job.

The legal framework was written for human decision-makers. It assumes intent, bias, and accountability that can be traced to individuals. Algorithmic decision-making distributes responsibility across systems, data pipelines, model designers, and deployment decisions — none of which maps cleanly onto existing law.

Judge Richard Seeborg is scheduled to hear arguments on July 23. The plaintiffs are seeking to pause the layoffs — which are scheduled to begin July 22 — while their claims go through private arbitration.

Not an American Problem

The Meta case is not isolated. It is part of a global pattern.

In China, courts have repeatedly ruled that replacing workers with AI is voluntary cost-cutting that does not justify mass layoffs. In one case, a Chinese court awarded a worker 260,000 yuan in compensation after his employer cited “AI impact” as grounds for termination.

In the United States, the Trump administration has ordered federal agencies to deprioritize disparate impact enforcement, arguing that its use undermines “meritocracy” — a policy shift that may affect how this case is ultimately litigated.

The legal vacuum is global. Algorithmic decision-making in HR is widespread — screening resumes, evaluating performance, making termination decisions. The technology is ahead of the law in every jurisdiction.

The Deeper Question

The Meta lawsuit raises a question that is becoming increasingly urgent: when an algorithm helps decide who gets fired, who is responsible?

Is it the engineer who built the system? The product manager who defined the metrics? The executive who approved the deployment? The HR leader who signed off on the process?

The plaintiffs’ argument is that Meta used AI to select layoff targets, but the company’s defense is that humans made the decisions. Both cannot be entirely true. And neither fully addresses the underlying problem: the legal framework was not designed for decisions made by systems that no single person understands fully.

The 26 plaintiffs are asking the court to pause their terminations until their claims can be arbitrated. But the broader question — who is accountable when AI decides who gets fired — will not be resolved by this case alone.

It is a question that courts, legislatures, and companies will be wrestling with for years to come. Not because Meta is unique. Because Meta is just the first.