AI, Protected Leave, and Reduction in Force: Employer Lessons from Does 1 Through 26 v. Meta Platforms, Inc.

PDF

On July 13, 2026, twenty-six current and former Meta employees filed a Complaint for Injunctive and Declaratory Relief in Aid of Arbitration and for Damages (the “Complaint”) in the U.S. District Court for the Northern District of California, together with a Motion for Preliminary Injunction (“PI Motion”) and an Application for Temporary Restraining Order (“TRO”). Plaintiffs allege that Meta used a “constellation of internal artificial-intelligence systems” to “score, rank, and select” employees for the mass reduction in force that Meta began announcing on May 20, 2026 (the “RIF”), rather than relying on “the considered judgment of managers who knew the work.” The Complaint alleges that these artificial-intelligence (“AI”) tools draw on inputs that, “by design, cannot be accumulated by an employee who is on protected medical or family leave, or whose output is reduced by a disability.” Plaintiffs further allege, on information and belief, that Meta did not “neutralize those inputs” for protected leave and did not pause the system for an individualized, leave-neutral review. As a result, Plaintiffs claim employees who took protected leave were “disproportionately selected for layoff” based on scoring that not only ignored their leave but effectively penalized them for exercising statutorily protected rights. Whatever its outcome, the lawsuit offers employers practical lessons on the precautions to take before using AI tools in critical employment decisions such as hiring, discipline, and termination.

Plaintiffs’ Allegations

According to the Complaint, the Plaintiffs are comprised of twenty-six current and former Meta employees, each of whom (1) “Meta selected for separation in the RIF” and (2) within the twenty-four months preceding the RIF, “took, requested, or was approved to take statutorily protected leave;” attempted to take protected leave but had interference; “or requested or received a reasonable accommodation for a disability.” Plaintiffs worked in California, Illinois, Washington, New York, the District of Columbia, Pennsylvania, and Florida. Further, four Plaintiffs hold employment-based, nonimmigrant visas sponsored by Meta.

The Complaint alleges that Meta used a “constellation of internal artificial-intelligence-assisted systems”—including “Metamate,” an internal assistant; employee-trained “second brain” agents that “ingest each employee’s communications and documents to replicate the employee’s output;” “algorithmic productivity scoring drawn from keystroke, screen-content, mouse, browser-history, messaging, and email data captured continuously from Meta-issued devices;” “internal dashboards displaying employee-level AI-token consumption;” and “AI-assisted performance-review and calibration tools that have substantially supplanted the prior manager-driven calibration process”—to select employees for the RIF’s termination list. On information and belief, Plaintiffs allege that the RIF selection inputs included each employee’s “rolling twelve-month performance rating; the Year-End 2025 calibration outcome; an ‘AI-native’ rating drawn from AI-token-usage dashboards; output volume and code-commit data; and manager-sponsorship and roadmap-adjacency signals.” According to the Complaint, numerous employees noticed that their scores had “dropped dramatically upon returning from vacation and protected leaves,” and that “there was no effort made by Meta to ‘pause’ a person’s scoring when they were away from work based on excused, and even legally protected, absences.” Plaintiffs allege that Meta “did not neutralize any of the foregoing inputs for protected leave.” As an illustration, an employee on extended protected leave cannot, by Meta’s alleged system’s design, “accumulate code commits, AI-token consumption, output volume, or ‘strategic alignment’ signals” at the rate of a peer who was not out on protected leave. They further allege that “Meta did not adjust for this, did not exclude protected-leave time from any throughput metric, and did not flag protected-leave-takers or accommodation-seekers for individualized human review.”

Notable accounts alleged in the Complaint and supporting declarations include: (1) a scientist selected for the RIF while on approved pre-birth pregnancy leave (selected just two days before she gave birth); (2) a manager whose performance review documents “that his demotion followed his return from medical leave” (selected for the RIF sixteen days into a second medical leave); (3) an engineer whose manager “tied his lowered rating directly to the ‘broken time’ when an injury prevented him from working;” and (4) a researcher who received the first lowered performance review (“Meets Most”) rating of her career a week after disclosing a diagnosis and requesting accommodations.

The Complaint asserts twenty-one causes of action. The federal claims are FMLA interference (use of protected leave as a negative factor) and retaliation under the Family and Medical Leave Act (“FMLA”); disparate treatment and disparate impact under Title VII and the Pregnancy Discrimination Act; violation of the Pregnant Workers Fairness Act; and disability discrimination, failure to accommodate, and retaliation under the Americans with Disabilities Act (“ADA”). Plaintiffs also assert state and local claims under the laws of California, Washington, New York, the District of Columbia, Florida, Illinois, and Pennsylvania, as well as claims for declaratory relief and for preliminary and permanent injunctive relief in aid of arbitration to maintain status quo.

In the TRO and the PI Motion, Plaintiffs identified several harms they argued were irreparable: (1) loss of employer-sponsored health insurance during active medical need; (2) loss of protected leave time, the purpose of which is “to heal, to care for family, and to have protected time away from work;” (3) forfeiture of unvested equity; (4) loss of employment visas and immigration status for the four visa-holding Plaintiffs; and (5) the inability to “undo” the RIF once finalized.

History of the Case

Because Plaintiffs signed Mutual Arbitration Agreements with class, collective, and representative-action waivers, each Plaintiff must pursue the merits individually, and Plaintiffs have initiated arbitration with the American Arbitration Association of their cases. In court, Plaintiffs seek only relief to preserve the status quo pending arbitration.

Meta opposed both the TRO and the PI Motion, supported by declarations from personnel involved in planning the RIF, including its Director of Human Resources Business Partner Enablement. Meta unequivocally denies using AI to decide which employees to include in the RIF. Rather, Meta asserts that selection decisions were “made by human business leaders based on documented, neutral criteria,” which did not include leave status, leave history, disability status, or accommodation requests, and it listed those criteria in its filings. Meta also characterizes Plaintiffs’ requested relief as an “attempt to eviscerate the at-will employment doctrine, evade the binding individual arbitration agreements that Plaintiffs voluntarily signed, preserve highly valuable unvested equity to which they are not entitled, and conscript this Court into supervising Meta’s legitimate business decisions.”

On July 16, 2026, the Court heard arguments on the TRO, and on July 17, 2026, the Court entered an order denying it. The Court found that Plaintiffs had raised “serious questions going to the merits” but had not demonstrated irreparable injury at this juncture, noting that “Meta’s declarations that it did not use AI in any way with respect to the RIF are unequivocal” and that discovery in arbitration “will be necessary to test the credibility of Meta’s declarants.” The Court held that losing employment, health coverage, protected leave, and unvested equity are not irreparable harms because each can be remedied in arbitration; however, the Court held “[l]oss of immigration status is different.” Because the visa-holding Plaintiffs face “an imminent threat of removal from the United States, absent finding new employment in the next sixty days,” the Court found that these conditions “likely constitute irreparable harm and may warrant injunctive relief.” Lacking sufficient information on why those Plaintiffs were selected in the RIF, the Court ordered Meta to submit a declaration explaining how and why the four Plaintiffs with Meta-sponsored employment visas were selected for termination. The Court also stated that it may reconsider its determinations based on additional evidence “regarding whether and how AI was used in an improper manner to cause the RIF of the plaintiffs.”

Both parties submitted additional declarations regarding the four visa-holding Plaintiffs, and according to the public docket, a hearing was held on the PI Motion on August 24, 2026. The Court has not yet entered a ruling on this Motion. While this case is still at an early stage and in arbitration, the filings so far offer lessons for employers.

Lessons Learned

1. Scrub Your Data for Hidden Biases.

The Meta case shows that a metric can look neutral and still penalize employees for exercising protected rights. If your company uses AI tools or data-driven metrics to evaluate employees, make sure those metrics account for protected leave. Four safeguards can help:

• Neutralize protected leave in the data. Exclude protected-leave periods from the measurement window, or prorate metrics so employees are measured only against the time they actually worked.
• Be especially careful with volume and activity metrics. Keystrokes, code commits, AI-token usage, and other “activity” measures reward simply being present. As asserted in the Complaint, an employee on leave cannot generate these signals at the rate of a continuously present peer.
• Find out why a score dropped. When an employee’s metrics fall suddenly, find out why before acting on the number. A drop that matches a leave period reflects absence, not performance. One director in the case attested that Meta’s dashboards “failed to distinguish employees’ leave from lack of engagement.”
• Check earlier ratings, too. Several Plaintiffs allege they received the first below-par rating of their careers shortly after returning from protected leave. If a prior rating was affected by leave, it will carry that bias into any later decision that relies on it, including a RIF.

These safeguards help keep your data measuring performance rather than absence. They also put your company in a stronger position to defend its decisions if they are challenged.

2. Plan for Visa-Holders and Other High-Risk Groups.

The Meta case shows that certain employees can make an ordinary layoff harder to defend. The court rejected most of the Plaintiffs’ claimed harms. It held that losing a job, health coverage, protected leave, or unvested equity can be remedied later in arbitration. However, the Meta Court held that “[l]oss of immigration status is different” and the court found their situation “likely constitute[s] irreparable harm.” It then ordered Meta to explain “how and why” it had selected those employees.

Visa holders present the strongest case for emergency relief. Also, other employees carry heightened risk of leave, discrimination, and retaliation claims: (1) employees who are out on protected leave at the time of the RIF, (2) employees who are out on protected leave in the time period prior to the RIF, (3) pregnant employees, and (4) employees with accommodations. Here are some measures to take to prepare for terminating from these groups:

• Have a selection explanation ready. For each employee in these groups, document the reason for selection before notices go out. This is exactly the record the court required Meta to submit.
• Review separation timing. Plaintiffs allege that their separation dates cut approved leave short or fell just before equity vesting dates. Check these dates and adjust them where appropriate.
• Bring in immigration counsel early. Before the RIF is announced, coordinate on grace periods, transfer options, and the timing of any notice to sponsored employees.
• Remember that arbitration does not keep you out of court. Meta’s arbitration agreement expressly allowed court actions for temporary or preliminary relief “to maintain the status quo pending arbitration.” Review the carve-outs in your own agreements, and be prepared for high-risk employees to use them.

These measures will not eliminate the risk. However, they will help you keep a well-documented and defensible record in the event one of these employees seeks emergency relief after a RIF.

3. Build in Individualized Human Review, and Document It.

Human review is a defense only if the review actually happened and the employer can show what it involved. Meta’s response to the lawsuit rests on that kind of record. Its declarant stated that RIF selections were “made by human business leaders based on documented, neutral criteria,” and Meta listed those criteria in its filings.  At the TRO stage, the Court found that “Meta’s declarations that it did not use AI in any way with respect to the RIF are unequivocal.” However, the Court did not treat those declarations as the final word. It found that Plaintiffs had raised “serious questions going to the merits” and that discovery in arbitration “will be necessary to test the credibility of Meta’s declarants.” Whether or not AI plays a role in your selection process, the following procedures will help you build a record to defend your selection decisions:

• Write selection criteria first. Set selection criteria before anyone is chosen, and avoid any criterion that cannot be explained in plain terms. Meta’s ability to identify specific, documented criteria was no doubt helpful to its defense.
• Involve managers who know the work. Decisions made only at a senior level, with no input from the employee’s own manager, invite the argument that data, not judgment, drove the outcome. Some of the more damaging Plaintiff accounts were ones where they were told that their managers did not have any input in the selection process.
• Add a second review for high-risk groups. Send everyone who is on leave, recently returned, or under an accommodation through a separate review step. This can help you take a second look to ensure that their leaves or accommodations did not affect the metrics or ratings behind their selection, and also makes the decision to include them easier to defend.
• Record the reasons. For each selection, document who decided and which criteria applied and why. Make sure to keep this record for any potential discovery. This is the type of information the Court ordered Meta to provide for the four visa-holding Plaintiffs.
• Avoid rubber-stamping. If reviewers approve the tool’s output without looking at anything else, courts are likely to treat the decision as the tool’s decision, not the reviewer’s decision.

These procedures will not prevent an employee from seeking a TRO or preliminary injunction. However, it helps to be prepared to defend RIF decisions.