If your company uses productivity data, performance metrics, or AI-assisted tools in workforce decisions, a new lawsuit against Meta Platforms, Inc. deserves your attention. Not because of what it proves (nothing, yet), but because of the question it raises: what are your “neutral” metrics actually measuring?

On July 13, 2026, twenty-six current and former Meta employees filed suit in the Northern District of California, alleging that Meta’s May 2026 reduction in force—which affected roughly 8,000 employees—unlawfully targeted people who had taken protected leave, were pregnant, or had disabilities or accommodations. The plaintiffs allege Meta used a “constellation” of AI-assisted systems and data-driven metrics including productivity data, AI-tool usage, performance ratings, and code commits to score and rank employees, and that those systems treated protected time away from work as reduced productivity.

The theory is simple. An employee who takes twelve weeks of protected leave necessarily generates fewer code commits, less output, and less workplace activity than a colleague who worked straight through. If the system measures total output without adjusting for the absence, legally protected leave quietly becomes a negative employment factor. One plaintiff alleges her AI-adoption score dropped after leave because the system treated her time away as a gap rather than removing it from the measurement period.

Meta disputes all of it. The company says human business leaders, not AI, made the selections using documented, neutral criteria, and no court has found otherwise. On July 17, Judge William Orrick denied the employees’ request for emergency relief, though he noted the case raised “serious questions” that discovery will need to sort out.

So why pay attention to unproven allegations? Because the underlying problem doesn’t depend on how this case ends. A metric doesn’t have to mention disability, pregnancy, or FMLA status to create discrimination risk. The FMLA prohibits using protected leave as a negative factor, and facially neutral practices can be challenged for disparate impact. The algorithm may never “know” an employee took leave. But if the absence depresses the inputs, the employer may be penalizing the leave anyway. The technology is new; this employment law problem is not.

California employers have an extra reason to watch. The plaintiffs expressly invoke California’s automated-decision-system regulations, which took effect October 1, 2025. Those rules are no longer just a compliance to-do—they are now showing up in complaints.

And one uncomfortable lesson: “human in the loop” may not be enough. If a manager simply approves a ranking built on flawed inputs, the human review adds little. What matters is whether the reviewer understands how the ranking was generated, whether protected absences were accounted for, and whether the authority to override it is real and documented.

What should employers do now?

• Examine the inputs, not just the outputs. Know what each metric measures and whether protected leave, accommodations, or disability-related limitations affect an employee’s ability to generate it.
• Normalize protected absences. Review productivity, utilization, and activity metrics to confirm employees on protected leave had an equal opportunity to accumulate them.
• Test before you decide. Before finalizing a RIF, assess whether the methodology disproportionately affects protected groups and investigate unexplained disparities.
• Make human review meaningful, and document it. Decision-makers should understand the criteria, be able to question or override recommendations, and leave a record of that review.
• Preserve AI-related records. Assume inputs, rankings, model outputs, and related communications will become evidence if a decision is challenged.
• Get HR, legal, and technology in the same room before systems influence consequential workforce decisions.

The first wave of AI employment litigation was about hiring—whether automated screening disadvantaged applicants. This case is about the people already on your payroll. A productivity metric may also measure time away on protected leave; an AI-adoption score may simply measure whether someone was present to use the tools. None of that means employers must abandon workforce analytics. But as employment decisions become increasingly data-driven, employers must be prepared to defend not only what decision they made, but how the data produced it.