Meta's AI Layoff Scorecard Can't Count Leave

Twenty-six Meta workers just filed the first major U.S. lawsuit accusing a Fortune 500 company of using AI systems to pick who gets laid off — and the complaint says the algorithms penalized anyone whose keystrokes, token usage, or productivity scores flatlined during protected leave.

The suit landed late Monday in Oakland federal court, one week before 8,000 notified separations begin on July 22. Plaintiffs want a preliminary injunction blocking those cuts while an independent auditor recomputes scores with leave- and accommodation-neutral inputs.

What the complaint alleges:

• Meta built its termination list with a "constellation" of internal AI tools — not manager judgment — including the Metamate LLM, employee-trained "second-brain" agents, keystroke and screen monitoring, and dashboards ranking workers as "AI Native," "AI First," or "AI Enabled," per Ars Technica's review of the filing.
• Those metrics "by design, cannot be accumulated by an employee who is on protected medical or family leave, or whose output is reduced by a disability," the complaint says — yet Meta allegedly never neutralized the inputs or paused the system for individualized review.
• All 26 anonymous plaintiffs took protected leave or received disability accommodations in the prior 24 months. About half took pregnancy or caregiving leave; eight women were on maternity leave, four men on parental leave, per AP.
• One scientist was selected while on approved pre-birth pregnancy leave — notified two days before she gave birth, The Guardian reported. Another manager was let go 16 days into medical leave.

Meta's defense:

"These claims lack merit and are not based on facts," a spokesperson told CBS News. "Workforce management and organizational decisions were and are made by people, not AI."

The company has not explained how human reviewers overrode telemetry that structurally zeros out during leave.

Why the timing stings:

• Chief People Officer Janelle Gale announced the ~10% cut in May to "run the company more efficiently" and offset other investments — the same month Meta reported record revenue and pledged $125 billion to $145 billion in 2026 AI capex, the complaint notes.
• Meta's employee-monitoring program — designed to train AI on keystrokes, mouse activity, browser history, and messages — drew backlash from 1,600+ staff and was paused in June, The Guardian reported. The lawsuit says workers got no opt-out on some teams.
• Mark Zuckerberg defended the program internally: "The AI models learn from watching really smart people do things," per The Information reporting cited in the complaint.

The legal frontier:

• Reuters characterized the suit as the first against a major U.S. company challenging alleged AI use in layoffs, per Ars Technica's reporting.
• Plaintiffs invoke federal laws — FMLA, ADA, Pregnancy Discrimination Act, Pregnant Workers Fairness Act — plus California's FEHA rules on automated decision systems, effective October 2025, which bar disparate-impact discrimination from AI-assisted termination tools.
• The complaint leans on disparate impact doctrine — facially neutral scoring that falls harder on women who take pregnancy leave — even as the Trump administration has ordered federal agencies to deprioritize that theory, AP noted. Private suits and state statutes still apply.
• Plaintiffs cannot pursue a class action — Meta's arbitration agreements block it — but seek court-ordered job preservation pending individual arbitration.

What's at stake July 22:

Plaintiffs' lawyers warn separations trigger irreversible harm: health coverage during postpartum recovery, forfeited unvested equity, extinguished leave rights, and immigration consequences for visa holders.

They want Meta to preserve all models, weights, and selection data — and let an outside auditor recompute who should actually be on the list.

Convina's view: Meta spent the spring telling Wall Street that 8,000 cuts would fund an AI-native workforce — then allegedly ranked that workforce on AI token dashboards that cannot tick upward during maternity leave. Whether a judge agrees is secondary: every enterprise now running productivity telemetry on AI adoption has a July 13 receipt that the metrics you optimize for will become the evidence against you. The human-in-the-loop disclaimer is not a legal strategy. It's a press release.