The nursery walls were already painted a soft, muted green when the notification arrived. It was May, the kind of afternoon where the air feels heavy with expectation. For one Silicon Valley scientist, the expectation was literal. Her baby was due in forty-eight hours. Her approved pre-birth pregnancy leave had finally begun, offering a brief, legal pocket of peace before the chaos of labor.
Then her phone buzzed.
It was a notification from Meta, her employer. Her role was being eliminated. A day later, her water broke. Two days later, she gave birth while holding a termination notice instead of a maternity plan.
We have been conditioned to think of bias as a human vice—a sneer, a passed-over promotion, a whispered comment in a hallway. But bias in the modern corporate world looks much cleaner. It looks like a zero on a dashboard. It looks like an algorithm performing exactly as it was designed to do, unburdened by the messy reality of human biology.
A federal lawsuit filed in California by twenty-six current and former Meta employees pulls back the curtain on a terrifying new normal in corporate America. The plaintiffs allege that when Meta cut roughly 8,000 jobs—about ten percent of its workforce—the selection process was outsourced to a "constellation of internal artificial intelligence systems." These systems tracked keystrokes, mouse activity, code output, and the usage of generative AI tokens.
The math was simple. The machine looked for output. If you were on legally protected maternity leave, family leave, or disability accommodation, your output was zero.
The machine did not care why.
The Math of Absence
Imagine a digital assembly line. Every time a worker types a line of code, sends a message on an internal channel, or interacts with a corporate AI tool, a small digital counter ticks upward. To the software, these ticks are the pulse of the employee. High pulse means high value.
Now imagine you break your leg, or your spouse falls critically ill, or you give birth. The law grants you protected time away from the machine. You shut your laptop.
But the algorithm does not pause. It continues to scan the team, averaging the metrics, ranking every individual from top to bottom. While a worker is recovering in a hospital bed or holding a newborn, their digital pulse drops to zero.
Consider what happens next: the algorithm evaluates the past quarter. It sees a flatline. It does not possess the context to understand that the flatline is legally sanctioned. It simply registers a sudden drop in performance, re-ranks the spreadsheet, and moves the flatlining name to the bottom of the list.
According to the lawsuit, one Meta director who joined the litigation admitted he reviewed these AI-adoption dashboards almost daily. The software made no distinction between a worker who had checked out emotionally and a worker who was recovering from major surgery. Another plaintiff, an engineer, watched his performance rating crater because a manager tied his evaluation directly to the "broken time" he missed while recovering from an injury.
Meta has publicly denied the claims, stating that workforce management decisions "were and are made by people, not AI." But the lawyers representing the workers argue that a human manager rubber-stamping a list generated by a biased mathematical model is a distinction without a difference. The algorithm creates the reality; the human merely hits 'send.'
The Disparate Ghost in the Machine
This is the modern face of what legal scholars call "disparate impact." It is an old legal doctrine, born out of the civil rights battles of 1971, which dictates that an employment practice cannot disproportionately harm a protected group, even if the policy sounds neutral on paper.
An algorithm that measures pure, uninterrupted keystroke volume sounds neutral. It does not explicitly hate women. It does not harbor a grudge against the disabled.
But women disproportionately take pregnancy and caregiving leave. Workers with chronic illnesses or sudden injuries disproportionately require accommodations that alter their daily metrics. When you build a system that rewards nothing but raw, unbroken digital friction, you build a system that systematically purges parents and the sick.
The irony is thick enough to choke on. Meta's leadership has been aggressively pushing to train its AI models on its own employees' behaviors. "The AI models learn from watching really smart people do things," the chief executive told staff in an internal meeting. The workforce was monitored, tracked, and scraped to build the next generation of software.
The reward for being the blueprint? Getting sorted out of existence by the very metrics meant to mimic you.
The Irreversible Cost
Corporate restructuring is often spoken of in the language of physics—"streamlining," "right-sizing," "adjusting velocity." It sounds clean, like water moving through a pipe.
The ground-level reality is messy, bloody, and terrifying.
The twenty-six plaintiffs are currently fighting for an emergency injunction to halt their terminations, which are scheduled to take effect on July 22. Their legal counsel points out that once these separations are finalized, the damage cannot be undone by a future court settlement.
When the laptop goes dark, the employer-subsidized health insurance vanishes. It vanishes during postpartum recovery. It vanishes in the middle of active medical treatments for disabilities. Unvested equity—the financial security tech workers trade their nights and weekends for—is instantly forfeited. For immigrant workers on specialized visas, the algorithm's calculation triggers a ticking clock toward deportation.
Technology has evolved faster than the human guardrails meant to keep it ethical. We have built tools capable of analyzing billions of data points in a second, yet those same tools cannot comprehend the basic human necessity of taking twelve weeks to care for a new life or heal a broken body.
The suit isn't just a fight over twenty-six jobs in Oakland. It is a referendum on whether corporate America can use the black box of automation as a shield against the law. If an algorithm makes the discriminatory decision, and a human leader merely executes it, who goes to court?
For now, twenty-six people are waiting on a federal judge, while in a quiet, green-painted nursery, a new mother logs into a portal that no longer recognizes her name.
