When Meta announced plans to cut 10% of its workforce earlier this year, the official line was all about efficiency. The company wanted to remake itself into an "AI-first" business. But a major lawsuit filed in a California federal court reveals a dark side to this transition.
A group of 26 current and former employees has sued Meta, claiming the tech giant relied on a complex web of internal AI tools and activity trackers to pick who got fired. The result? Workers who took protected medical, parental, or family leave were systematically targeted.
This isn't just about one company. It's a wake-up call for the entire tech sector. When you outsource human resource decisions to algorithms, you don't eliminate bias. You just automate it.
The Cold Metrics of a Digital Sacking
If you've worked in corporate tech lately, you know the pressure to perform is constant. But Meta allegedly took this to a whole new level.
According to the 71-page complaint filed in Oakland, California, Meta didn't rely on the careful judgment of managers who actually knew their team's daily contributions. Instead, they handed the keys to a suite of automated monitoring tools.
These systems tracked everything. Keystrokes. Screen activity. Browser history. Messages and emails. Even voice and location data. On paper, this was supposed to build a objective picture of worker productivity. In reality, it created a system where if you aren't actively clicking, you don't exist.
The problem with this approach is obvious. If you're on approved leave to care for a newborn or recover from a major surgery, your metrics go to zero. The algorithm doesn't care that you have a legal right to that time off. It just sees a flatline on a dashboard and flags you for removal.
Inside the Black Box of Metas Layoff Machine
To understand how this happened, you have to look at the specific tools Meta put in place. The lawsuit paints a picture of a hyper-monitored workspace driven by several key systems.
- Metamate: An internal AI assistant built to help employees with daily tasks but used to measure engagement.
- Second-Brain Agents: Custom-trained AI agents designed to handle specific engineering workflows, tracking who contributed what.
- AI Token dashboards: Metrics showing how much computational power and AI resources an individual engineer consumed.
- Activity-Monitoring Software: Programs that tracked active mouse movements, keystrokes, and software usage throughout the day.
Imagine how this plays out in real life. One plaintiff in the lawsuit took approved pre-birth maternity leave. Because she was away, her token-usage scores plummeted. Her activity tracking showed nothing but silence. To a manager, she was a valued colleague taking a standard, legal break. To the automated selection tool, she was a dead weight dragging down the team's average.
Another employee disclosed a serious medical condition that Meta's own medical providers had approved for leave. But his manager explicitly discouraged him from taking it. The manager warned that the system would flag his absence and guarantee his spot on the layoff list. When the employee brought up local family leave protection laws, the manager reportedly answered bluntly, "This is Meta. If the company wants something, they do it."
Why the Math of Disparate Impact Spells Trouble
Meta's legal defense is simple. A spokesperson stated that the claims lack merit, asserting that "workforce management and organizational decisions were and are made by people, not AI."
But that defense misses the point of modern civil rights law.
Under Title VII of the Civil Rights Act of 1964, you don't have to prove that a company actively hated a specific group of people to win a discrimination case. You just have to prove "disparate impact." This legal doctrine, established in the landmark 1971 Supreme Court case Griggs v. Duke Power Co., says that if a neutral-looking policy disproportionately hurts a protected group without a strict business necessity, it's illegal.
By tracking daily digital output and using those unadjusted scores to build a termination list, Meta's process fell heavily on women and employees with medical conditions. Women disproportionately take pregnancy and caregiving leave. People with chronic illnesses or disabilities take medical leave.
If you use an algorithm that scores people on continuous, uninterrupted digital activity, you will naturally select women and disabled workers for layoffs. You can't just throw up your hands and blame the computer.
The Human Cost of Managing by Dashboard
Tech executives love to talk about data-driven decisions. They want clear, clean numbers. But human lives are messy, and they don't fit into neat spreadsheet columns.
When an employee is fired, the consequences are immediate and severe.
- Lost Health Coverage: For pregnant workers or those in active medical treatment, losing employer-subsidized health insurance is a disaster.
- Equity Forfeiture: Years of unvested stock options disappear overnight, wiping out a huge chunk of expected compensation.
- Immigration Crises: For foreign workers on H-1B visas, a sudden layoff triggers a ticking clock to find a new job or face deportation.
The plaintiffs' lawyers are fighting for a preliminary injunction to stop these terminations and restore the workers' status while the legal battle goes to arbitration. They argue that once a worker is cut, the damage is done. You can't easily undo the trauma of losing your home, your visa, or your healthcare during postpartum recovery.
How to Build a Fair Process That Actually Survives a Legal Audit
If you run a business or manage an HR department, you might be tempted to look at Meta's situation and think, "We just won't use AI." But that's not realistic. Automation is here to stay.
The real lesson is that you must build guardrails into your automated tools. If you use performance-tracking systems, you have to establish a clear, manual review phase that accounts for protected absences.
First, pause the data collection for anyone on leave. If an employee is on FMLA, their metrics must be frozen or adjusted to reflect their actual working hours. You cannot let a zero-productivity score drag down their overall evaluation.
Second, don't keep the mechanics of your evaluation tools a secret. Transparency is your best defense. If employees know how they are being measured, they can point out errors or discrepancies before those errors turn into a wrongful termination lawsuit.
Finally, never let an algorithm make the final call. AI should only provide data points, not decisions. Every single termination list needs to be audited by human HR professionals who understand employment law and can check for disparate impact.
If you rely solely on numbers, you will eventually find yourself in a courtroom trying to explain why your computer system fired a pregnant woman on approved leave. And "the algorithm made us do it" is not a defense that works in front of a federal judge.
Take a hard look at your internal tracking systems today. Ensure your metrics are adjusted for leaves of absence, and make sure human eyes review every single cut. Don't wait for a lawsuit to show you where your data is broken.
