Twenty-six staff at Meta who have been just lately laid off made a daring declare in a lawsuit final week: They alleged the corporate had used AI to pick individuals for layoffs, disproportionately impacting individuals who had taken go away or these with disabilities. The lawsuit recognized plenty of elements that the plaintiffs claimed had influenced the layoff choice course of, from an inner AI assistant to dashboards that tallied staff’ AI token utilization.
The choose within the case requested the corporate to clarify its choice to terminate 4 of the plaintiffs whose immigration standing was in danger, after denying the plaintiffs a brief restraining order—which might have blocked Meta from finishing the layoffs.
In a filing on Thursday, Linh Doan, an organization HR chief defined its rationale for choosing these staff for termination—all of whom had visas sponsored by Meta—providing perception into how the corporate conducts layoffs. Doan, the director of HR Enterprise Companion Enablement, repeatedly emphasised that these choices have been made by “human enterprise leaders” and that AI was not concerned within the choice course of, disputing the claims put forth by the plaintiffs. She additionally denied that the leaders thought of—and even had perception into—whether or not an worker was on go away, claiming layoff standards “didn’t embrace go away standing, go away historical past, incapacity standing, lodging requests, or some other protected attribute.”
Doan revealed that the corporate first recognized the elements of the group that may be impacted by layoffs—right down to a “cohort” of staff with a selected degree and title—after which decided a enterprise rationale for why the group needs to be included within the layoffs. Then leaders needed to give you “choice standards” to guage the workers in that group.
“The choice standards needed to be goal and tied to the enterprise rationale,” Doan mentioned within the submitting. “For instance, if the enterprise rationale was to retain high expertise inside a specific group, the choice standards would possibly determine staff with decrease efficiency scores or decrease historic efficiency scores.” Past efficiency, the choice standards may additionally embrace any of the next components: job profile or degree, tenure, location, specialised abilities or expertise, in addition to what the corporate describes as spans and layers. (Spans refers to what number of direct studies a supervisor has and layers captures the place an worker sits within the group relative to CEO Mark Zuckerberg.)
In accordance with Doan, the choice standards for a cohort was finalized earlier than any particular person worker was evaluated.“Resolution-makers didn’t have the flexibility to deviate from these standards,” she claimed.
The declaration additionally offered particular explanations for why every of the 4 staff with visas have been laid off—and in three of these circumstances, the workers have been chosen for layoffs for performance-related causes. For one of many plaintiffs on this group (described within the lawsuit as “Doe 4”), Doan claimed that the choice course of for layoffs concerned “two goal choice standards utilized in sequence,” which included job degree and historic efficiency. The worker in query was on parental go away when he was laid off and had acquired a “Persistently Met Expectations” ranking in his year-end 2025 assessment.
“Meta maintains a high-performance tradition with seven attainable efficiency scores,” Doan mentioned. “Persistently Met Expectations” is the bottom ranking that doesn’t fall beneath expectations. The choice to pick staff assembly these standards was a part of a broader effort to cut back headcount at this degree whereas retaining higher-performing staff and transitioning to a smaller, talent-dense group.”
Whereas Meta insisted that AI was not used to really make layoff choices, Doan’s declaration didn’t contact on whether or not an worker’s AI utilization may affect their efficiency scores, given these scores have been a key criterion for sure layoffs. Previous reporting has recommended that Meta does, in truth, take into account staff’ AI adoption in its efficiency assessment course of, and that the corporate has began utilizing a software referred to as Checkpoint. (When reached by Quick Firm, Meta was not instantly accessible for remark.)
This lawsuit has additionally supported these claims, citing allegations from a director on the firm who mentioned “worker AI adoption was going to be a core evaluation metric” as a part of the Checkpoint program, the small print of which weren’t clearly disclosed till after the Could layoffs have been introduced. He claimed to watch AI utilization dashboard “nearly every day” and that these dashboards didn’t account for workers being out on go away. This declare was echoed by a number of different plaintiffs, who mentioned their AI utilization scores had dropped considerably whereas out on go away.
As Quick Firm has previously reported, many staff throughout company America report dealing with repercussions at work after they return from go away—and even whereas they’re nonetheless on go away. Current circumstances additionally recommend that taking parental go away can influence efficiency evaluations: Joanne Barela, a former human capital advisor who labored at Deloitte, introduced a case through which she claimed the corporate had penalized her in efficiency evaluations after she went on parental go away. In 2024, Microsoft settled a case with the California Civil Rights Division for over $14 million, after the company discovered that staff had allegedly confronted retaliation and discrimination for taking go away, together with decrease bonuses and efficiency scores.
Because the lawsuit in opposition to Meta suggests, there are open questions concerning the unintended results of tying efficiency metrics to an worker’s embrace of AI, even when corporations usually are not explicitly outsourcing layoff choices to AI. For workers who might already face bias—be it working dad and mom or staff with disabilities—the penalties may very well be even larger.
