10,000 Workers Ranked, Protected Leave Made Invisible

Meta is being sued by 26 former employees who claim the company's AI systems targeted workers on parental and medical leave during layoffs.

The lawsuit centers on a specific failure. Performance-ranking tools evaluated all 10,000 dismissed workers simultaneously, but made no exclusion for protected leave status. The argument sounds straightforward—the company had a tool, the tool didn't account for a protected class, negligence therefore.

But the case turns on something the complaint doesn't quite articulate. The question is whether Meta's ranking architecture was even built to accommodate such exclusions at all. Performance-ranking systems operate on a premise that sounds neutral until you examine it closely—they work by establishing statistical validity across a population, with every worker entering the same calculation.

The moment you remove a subset of workers from the dataset—say, those on leave—you create a data gap. A statistical hole. The remaining workers now rank against a smaller, different population, and the algorithm's comparative validity fractures. This is where the architecture reveals its choice. Most ranking systems can technically be modified to exclude people. They're rarely designed with that possibility as a primary feature.

That's not a bug. It's an assumption built into the math itself. The requirement that everyone be rankable within the same system is treated as more important than the granular justice of context. In 2023, when Microsoft published its "Responsible AI Principles," it promised systems that could be "contextual and situational"—but the framing was careful, promising only that they should be, not that they are. The lawsuit assumes Meta failed to do something it could have done.

That's not a bug. It's an assumption built into the math itself.

But what if the question should be different? Does a system designed for universal, simultaneous ranking actually have the capacity to exclude? Or does that demand contradict its core function?

The 26 employees were ranked. They lost their jobs because they ranked poorly. The question now isn't whether Meta was careless. It's whether, by design, the tool made protected leave invisible—not as a failure of compliance. As the price of the tool's fundamental architecture working as intended.