Meta cut 8,000 jobs in May, announcing the layoffs while reporting first-quarter 2026 revenue of $56 billion. The company was not struggling. It was making a deliberate choice to replace human labour with AI infrastructure, committing up to $145 billion in AI spending this year alone.
To decide who would go, Meta did not rely solely on managers making calls about the people they knew. According to a 71-page lawsuit filed in federal court in Oakland, the company used a set of internal AI tools to score and rank employees. Those tools tracked things like keystrokes, AI token usage (how much employees used Meta's own AI assistants), code output, and productivity dashboards. The system then generated a ranked list, and that list became the layoff list.
The core legal problem is straightforward. Anyone on medical leave, parental leave, or disability accommodation was, by definition, not generating those data points. They were not at their keyboard. They were not running AI tools. The system had no mechanism to account for that. So workers on protected leave showed up near the bottom of the rankings, not because of their actual performance, but because they were legally entitled to be away from work.
One plaintiff, described in court documents, was a scientist on approved pregnancy leave who was terminated two days before her scheduled delivery date.
Meta's response is that all workforce decisions were made by people, not AI. That is technically possible, and it is the standard defence in these cases. Managers may have had final sign-off on names. But the lawsuit argues that when the list was generated by AI scoring and handed to managers, the human review was largely a formality. The question courts will eventually have to answer is: at what point does a manager reviewing an AI-generated list count as a human decision?
That question matters well beyond Meta. Across the tech sector, and now in industries like insurance, logistics, retail, and finance, companies are using AI-assisted tools to measure employee productivity. Some of those tools track application usage. Some count tasks completed. Some measure response times or client interactions. All of them, if they score volume of output over a set period, will systematically disadvantage employees who took time off for any legally protected reason.
The legal exposure here is also expanding. California's regulations covering AI in employment decisions took effect in October 2025. New York City already requires independent bias audits for any automated tool used in hiring or termination decisions. Colorado's law, which forces employers to assess whether their AI systems create discriminatory outcomes, takes effect this month. Illinois now prohibits using AI in employment decisions if the result disadvantages a protected group, even if that was not the intent.
The plaintiffs also allege that Meta failed to test its AI systems for bias, which would put it in direct violation of the California and New York City rules. That is a separate legal thread from the discrimination claim, and it is the kind of procedural failure that tends to be much easier to prove.
For any organisation running performance reviews, annual appraisals, or productivity tracking through software, the practical question is simple: does your system know when someone was on leave, and does it account for that when producing scores or rankings? If the answer is no, or you are not sure, the Meta case is worth watching closely. The legal standard is moving toward requiring that employers actively test for these outcomes, not just assume the software is neutral.