July 27, 2026
Meta Says Humans, Not AI, Made Its Layoff Decisions – but Questions Remain
Meta has denied that AI selected staff for its May 2026 layoffs, telling a US court that human leaders made the decisions using pre-set, neutral criteria. The sworn filing in the Meta AI layoffs lawsuit answers some questions but leaves others open – including whether AI shaped the performance scores behind the cuts.
Layoff selection decisions at Meta were made by “human business leaders based on documented, neutral criteria”, an HR leader at the software giant states in court filings.
A group of 26 former and current employees (Does 1–26) are suing the company, alleging it used a “constellation” of internal AI systems to select staff for its May 2026 layoffs, disproportionately affecting those on protected medical, family, or pregnancy leave.
The case, first reported by Reuters on 14 July 2026, appears to be the first against a major US company to challenge the alleged use of AI in conducting layoffs.
Judge William Orrick declined to block the layoffs, finding the workers had not shown the “irreparable harm” needed for an emergency order. But he ordered Meta to explain why four plaintiffs on work visas were selected. The immigration status of those plaintiffs – Does 4, 9, 15, and 26 – meant they had a limited window to find another employer to sponsor their work visas.
Meta’s response came in a sworn declaration filed on 23 July by Linh Doan, its Director of Human Resources Business Partner Enablement. What follows is Meta’s sworn evidence; the court has not ruled on whether it is accurate.
What Meta’s Court Filing Says About Its Layoff Criteria
Doan states that the layoff decisions were made by business leaders, using criteria determined before any individuals had been selected. This pre-determined criteria was “tied to a specific business rationale”, which the decision-makers were not allowed to “deviate from”.
This pre-set list consisted of: job level, historical performance, most recent performance rating, tenure, location, job function, specialised skills, and “spans and layers” (the number of direct reports a manager has, and an employee’s distance from Mark Zuckerberg in the hierarchy).
Decision-makers had to apply one or more of the “permissible criteria” to their “decisional unit”, factoring in the company’s objectives and its restructuring needs.
Did AI Select Staff for Meta’s Layoffs?
The declaration gives a clear response to the plaintiffs’ allegations on AI use: “AI did not select anyone for termination. AI did not assist anyone in making selection decisions. AI was not used to calculate performance scores, to rank employees, or to generate, influence, or recommend selection criteria.”
The statement also addresses an activity-monitoring tool, the “Model Capability Initiative”, which the plaintiffs claim Meta used within the selection process. Doan states that HR and business leaders could not access the tool’s data. The software was also not launched until 22 April 2026, after the selection process was “already well underway”, with some decisions already confirmed.
She also denies that AI-token usage, AI-adoption metrics, or “second brain” training data were part of the selection criteria, as the plaintiffs allege.
How Meta Says It Treated Employees on Leave
The declaration states that all employees in the selection process were assessed “uniformly” against the criteria. Employees who had taken any form of leave were judged by the same parameters as those who had not.
Doan explained that leave status, leave history, disability status, accommodation requests, and any other protected characteristic were not part of the permissible criteria. She also stated that decision-makers could not access information on employees’ leave status during the selection process.
Why Each of the Four Plaintiffs on Work Visas Was Selected
The declaration sets out an individual rationale for each of the four visa-sponsored plaintiffs, tying each decision to the pre-set criteria rather than to leave.
Two were on parental leave when notified. Doe 4, a Scientist in an individual contributor level 5 (IC5) position, was selected on job level and then historical performance. Meta says his leave “played no role”: he was chosen because his performance was lower than retained peers. His year-end 2025 rating of “Consistently Met Expectations” was, in Meta’s words, “the lowest rating that does not fall below expectations”.
The declaration adds that 13 employees in Doe 4’s cohort who were on leave, or had recently taken it, were not selected.
Doe 26, also an IC5 Scientist, was chosen on the same two criteria and notified less than two weeks after returning from parental leave. Meta says nearly 40% of her team was cut, and that no others affected in her cohort had any history of leave.
The other two were selected on skills and performance. Doe 9, a Designer, was cut for lacking specialised experience the business had come to need, despite an approved accommodation Meta says was irrelevant to the decision. Doe 15, an engineer, was selected on a “Met Most Expectations” rating, which Meta says signals below-expectations performance. Meta states the rating predated his health condition and leave request.
The Unresolved Gap: AI and Performance Ratings
Doan’s declaration is categorical that AI played no part in choosing who was cut. What it does not address is a step further back in the chain. Historical performance ratings were among the criteria used to select at least some employees, including two of the four visa holders. Yet the declaration is silent on whether an employee’s use of AI tools fed into those performance ratings in the first place.
That’s an important detail because the plaintiffs’ case rests partly on it. Their complaint alleges AI adoption had become a core workplace metric through an internal Meta programme known as Checkpoint, and that AI-usage dashboards did not pause when staff were on approved leave, so time away from work showed up as a drop in recorded activity.
Meta has denied that AI made the layoff decisions. It has not, in this sworn filing, addressed whether AI adoption shaped the performance scores that then helped decide who stayed.
What Happens Next in the Meta AI Layoffs Lawsuit
Following the 17 July ruling, the plaintiffs’ law firm argued that the case cannot be resolved on Meta’s account alone, since the company holds the internal records that would show how the layoffs were actually decided. It noted that the court had recognised the plaintiffs “were not in the rooms where it happened”, and that discovery “will be necessary to test the credibility of Meta’s declarants”.
The preliminary injunction hearing is set for 24 August, with Meta’s response to that broader motion due by 10 August.