Meta sued by 26 workers over AI-assisted layoff targeting

There is a first-of-its-kind employment case sitting in an Oakland federal court, and it matters more for what it will force into public discovery than for how it lands on the merits. Twenty-six current and former Meta employees, filing anonymously, allege the company used AI-assisted software to help pick who got cut in a mass layoff earlier this year, and that the resulting selection disproportionately hit workers with disabilities, workers on medical leave, and pregnant employees, as Reuters reported.

The named tools are the interesting part. The complaint calls out Metamate, Meta's internal large language model assistant; an employee-trained 'second brain' that tracked workers' communications and documents; and a productivity score built by scanning keystrokes, screen content, emails and browser history. Those signals feed a ranking that plaintiffs say punished anyone who missed work for a medical reason or to care for family. Meta's public line, also carried by CNBC, is that 'workforce management and organizational decisions were and are made by people, not AI.' That framing is going to get tested, because plaintiffs are asking the court to freeze the layoffs scheduled to take effect July 22 while the substance moves to private arbitration.

Why this matters for anyone outside Meta: this is the first suit against a major U.S. company squarely challenging the use of AI in a reduction in force. It also invokes the recently adopted California and New York City rules requiring bias testing of automated decision tools, which until now have been mostly theoretical. If plaintiffs get their preliminary block, every HR org that has been quietly wiring productivity-monitoring data into a headcount model has a very immediate problem to solve.

The honest caveat is that these are plaintiffs' allegations, filed anonymously, and Meta flatly denies that AI made the calls. What the reporting does not give you is which model or vendor built the scoring pipeline, whether a human review step existed between an AI-generated ranking and the actual cut, or what disparate-impact rate plaintiffs will put in front of the judge. Take the specifics, including the AI token usage input, as reported rather than settled.

The forward-looking piece is the vendor market and the regulators. Employment lawyers and HR-tech vendors selling explainable, independently bias-audited layoff tooling just got their flagship pitch, and the California and NYC automated-decision regulators just got a case they can point to.