Meta's AI Layoff Lawsuit Is a Warning for Every Company Scoring Employee Output

The lawsuit that turns HR data into a discrimination case

Twenty-six current and former Meta employees filed a federal discrimination lawsuit against the company on July 13, 2026, in the Northern District of California. The plaintiffs were part of the roughly 8,000 employees, about 10% of Meta's workforce, cut in the company's May layoff round. Separations for this group are scheduled to begin July 22.

The complaint alleges that Meta's layoff selection process ran on a mix of internal AI systems, including performance ratings, calibration scores, productivity and output metrics, so-called 'AI-native' ratings, and AI token-usage dashboards, along with keystroke and activity-monitoring data. According to the plaintiffs' lawyers, these inputs cannot be accumulated at a normal rate by anyone on protected medical or family leave, or by anyone whose output is reduced by a disability, simply because that is what the leave or the disability does to the underlying activity data.

About half the plaintiffs took leave for caregiving or pregnancy-related reasons. Eight are women who took maternity or pregnancy-related leave, four are men who took parental leave, and one is a woman who took leave to care for a family member and later bereavement leave. Meta told reporters the claims lack merit, stating that workforce management and organizational decisions were and are made by people, not AI. The company has not addressed the specific systems named in the complaint or confirmed whether any adjustment mechanism existed for employees on protected leave.

Why activity data quietly punishes leave and disability

The mechanics of the complaint matter more than the headline. The plaintiffs are not arguing that Meta built a biased algorithm on purpose. They are arguing that any scoring system built on activity volume, output counts, and token or tool usage will structurally undercount someone who was out sick, on leave, or working at reduced capacity because of a disability, whether or not the company intended that outcome.

That distinction matters because it maps directly onto disparate impact law. A policy or tool does not need discriminatory intent to be unlawful if it produces a discriminatory effect on a protected group and the employer cannot show the tool is job-related and consistent with business necessity. The lawsuit cites the Family and Medical Leave Act, the Americans with Disabilities Act, the Pregnancy Discrimination Act, and the Pregnant Workers Fairness Act, and argues Meta did not pause its layoff system for the individualized, leave-neutral review the law requires.

CNBC reported that the lawsuit describes Meta's inputs as a 'constellation' of internal AI systems, and that lawyers for the plaintiffs say these tools were never adjusted to exclude or normalize for approved absences before being used to rank people for termination. If accurate, that is not a hypothetical edge case. It is the predictable output of feeding raw activity data into a ranking system without a control for who was legally allowed to be less active.

The regulatory backdrop makes this riskier, not safer

Companies scoring employees with AI might assume the compliance bar has dropped given the current political climate. The Trump administration has directed federal agencies to deprioritize disparate impact enforcement, arguing the doctrine encourages the assumption that any demographic imbalance reflects discrimination, and the EEOC has dropped some discrimination cases as a result. The EEOC also quietly removed its AI-specific hiring guidance from its website in January 2025.

The Meta case shows why that retreat does not eliminate the exposure. Disparate impact liability is codified in Title VII of the Civil Rights Act of 1964, not created by EEOC guidance, and workers can still bring these claims directly, or route them through state law, even if the federal agency declines to pursue them itself. Illinois, Colorado, New York City, New Jersey, and California have each layered their own AI employment rules on top of the federal framework, several of which specifically apply the disparate impact standard to any tool that has the effect of discriminating on a protected basis, regardless of intent.

The Uniform Guidelines on Employee Selection Procedures, adopted jointly by the EEOC, DOL, DOJ, and OPM in 1978 and still in force, defines 'selection procedure' broadly enough to cover AI-driven scoring and ranking tools used in layoffs, not just hiring. Employers cannot shift that liability to a vendor either. If a third-party performance platform or monitoring tool produces the adverse outcome, the company using it remains legally accountable.

What to check before your next round of AI-assisted layoffs

This exposure is not unique to Meta, and it is not limited to layoffs. Any company using productivity scores, activity dashboards, AI-token consumption, calibration ratings, or algorithmically assisted performance rankings to inform hiring, promotion, or termination decisions is running the same risk, just at a smaller scale. A handful of practical steps separate a defensible process from a lawsuit waiting to happen.

• Run an adverse impact analysis on any scoring or ranking tool before it touches a real employment decision, and repeat it on a recurring cycle rather than once at rollout.
• Build an explicit exclusion or normalization step for employees on FMLA, disability, or parental leave before their data feeds into any automated ranking.
• Keep a human reviewer in the loop who can explain, in plain terms, why the tool ranked someone the way it did. A ranking system that cannot explain its own output is a liability, not a shortcut.
• Document the business necessity case for every metric in the model. If a metric cannot be tied to job-relevant output, its presence in a termination-adjacent score is very hard to defend later.
• Treat vendor-built scoring tools as your legal responsibility, not the vendor's. Ask for the vendor's own bias testing and keep records of what you asked and what you got back.

Sources