I was on a call when the notice hit—someone in the channel typed three words and the room went quiet. A scientist two days from giving birth saw her name on a list. You felt the sudden, peculiar sting of being evaluated by a process you couldn’t see.
I’ve read the complaint and followed the filings. I’ll walk you through what the 26 anonymous Meta employees say happened, what the company says in response, and why this case could change how workplaces use artificial intelligence.
A scientist two days from giving birth received a layoff notice.
The lawsuit, filed in the Northern District of California, says Meta didn’t rely on managers’ judgment when it cut roughly 8,000 roles in May—about 10% of the workforce—but on a network of internal AI systems. The complaint names an internal large-language model assistant called Metamate and several algorithmic metrics: keystroke and browser-history–derived productivity scores, AI token consumption logs, and AI-assisted performance-review tools.
According to the filing, those signals were fed into scores that “ranked and selected employees for inclusion on the list.” Plaintiffs say the system penalized people who had reduced output because they were on protected medical or family leave. The result, they claim, was disproportionate selection of workers on pregnancy, disability, or other protected leaves.
Can AI be used legally to make layoff decisions?
Short answer: sometimes, but not without legal risk. The plaintiffs are asking a judge to stop Meta from completing the layoffs on July 22 so they can pursue claims in private arbitration. Their argument: Meta’s employment agreements require individual arbitration for workplace disputes but do not strip the court of authority to grant temporary relief, like pausing the cuts while the claims are heard.
Meta told Gizmodo that “workforce management and organizational decisions were and are made by people, not AI,” and says the claims lack merit. But the company’s own filings and reporting by The New York Times spotlight its plans to pour huge sums into AI—reports put that spend in the ballpark of $100 billion (€93 billion)—which is the same fiscal pressure the lawsuit ties to automated headcount decisions.
Managers described being handed a list, not making one.
Several employees say their managers told them the termination list came from tools rather than human judgment. The complaint describes Metamate—an internal LLM trained on company emails and documents—plus algorithmic productivity scores and AI review assistance as the decisive inputs.
When systems weight keystrokes, email patterns, or token consumption, those signals can mirror availability, not value. The plaintiffs say Meta knew the system penalized people on protected leave and still relied on it, rather than pausing the automation for a neutral human review. The algorithm was a silent referee, and the company’s pause button, according to the suit, was never pressed.
How can employees dispute AI-based layoffs?
The workers are pursuing two tracks: stop the imminent terminations through a preliminary injunction and carry individual discrimination claims into private arbitration, as their contracts require. Their lawyers say an injunction is appropriate because arbitration clauses don’t remove the court’s ability to grant temporary relief. If the court agrees, Meta would be blocked from completing the July 22 cuts for those plaintiffs while arbitration proceeds.
This suit arrives after another bias allegation in February 2025.
Employees point to a pattern: earlier, a 2025 round of cuts that trimmed roughly 5% of staff was challenged as targeting older workers. Now similar concerns—this time focused on disability, pregnancy, and parental leave—are front and center. One engineer in the current complaint says he knew people who took paternity leave were laid off in that February round as well.
The legal theory is straightforward: if a scoring system fails to account for protected absences and uses metrics that correlate with being on leave, the system can have a disparate impact on protected classes. That’s where arbitration, injunctions, and discovery could reveal the data and rules behind the models.
The plaintiffs want the court to pause the July 22 layoffs so arbitration can proceed.
Their filing asks the judge to grant a preliminary ruling blocking the employment terminations while private arbitration claims move forward. If granted, the pause would give employees access to arbitration without immediate separation and preserve evidence—system outputs, token logs, manager notes—that could show the role AI played.
Meta denies the allegations and insists humans made workforce decisions. But the suit points to internal reliance on things like token consumption—an input tied directly to internal LLM use—as a factor in ranking. The scoring system turned workers into ledger entries, the complaint says, and those entries didn’t reflect protected leaves.
What is Metamate and how was it trained?
Metamate is described as an internal LLM assistant trained on employee communications and documents. Plaintiffs allege it was used alongside tools that analyze keystrokes, browser history, and email metadata. Those signals, the complaint argues, were assembled into productivity and performance scores that fed layoff decisions.
Think of popular providers and platforms—OpenAI, Google Cloud, internal telemetry systems—that companies use to build and monitor models; Meta has its own internal stack and token economy, and the suit highlights token consumption as a proxy metric that allegedly disadvantaged people who were away from keyboards for protected reasons.
This is more than a narrow workplace dispute. It tests how courts will treat automated decisions that touch legal protections, how corporations balance AI investment with labor risk, and what transparency employees can demand about data that determines their careers. I won’t pretend to have all the answers, but you deserve to know what the complaint claims and why it matters. Who watches the code when your livelihood depends on it?