Meta Lawsuit Puts AI Layoffs Under Scrutiny

Meta Lawsuit Puts AI Layoffs Under Scrutiny

A lawsuit brought by 26 workers, both current and former, at Meta says internal AI systems helped decide which employees would lose jobs in May, and the dispute centres on protected leave. The workers allege the systems unfairly singled out employees who had requested leave or taken it over the previous two years. Meta said people made the decisions, not AI.

Workers Describe How Meta Chose People to Lay Off

The workers filed the lawsuit Monday in the U.S. District Court for the Northern District of California over a cut affecting approximately 10% of the workforce.

The workers alleged that Meta used several internal AI systems to give employees scores and put employees in order, which helped determine who would lose their jobs. The workers said the tools used information about performance and employee activity to calculate scores used when choosing people for layoffs. According to the workers, those inputs included:

  • Ratings of employee performance and calibration scores
  • Measures of productivity and output
  • “AI-native” ratings
  • AI-token consumption

According to the lawsuit, employees on protected leave could not build up those measures in the same way as other employees, and employees with disabilities that limited their output could not either.

The workers alleged that Meta did not adjust the inputs for leave, and that Meta still considered employees for layoffs when employees had taken protected leave or requested accommodations. For companies making employment decisions with AI, the risk goes beyond whether the numbers are right. A score can accurately show how much time an employee was away and still give a misleading view of how that employee performed.

How Leave and Disability Figure in the Complaint

The workers pointed to several examples of leave allegedly changing how Meta treated them. The accounts are still allegations, and the lawsuit has yet to settle the differences between what the two sides say happened.

RoleLeave takenAlleged treatmentTiming
ScientistPregnancy leave before giving birthSelected to lose the jobWhile taking leave
ManagerLeave for medical reasonsGiven a demotion, then selected to lose the jobSeveral weeks into a second period of medical leave

An engineer said an injury meant the engineer could not work, and the engineer’s rating went down as a result.

The examples bring up questions about what performance scores actually tell an employer. How long someone works is not the same as how well someone works, and combining both in one score could work against employees who take leave. Using AI to come up with the score could also make it more difficult for workers to understand where things went wrong and to dispute the outcome.

The workers said Meta violated the Americans with Disabilities Act, along with the Family and Medical Leave Act, the Pregnancy Discrimination Act, the Pregnant Workers Fairness Act and Title VII of the 1964 Civil Rights Act. A preliminary injunction is what the workers are asking for, which would prevent Meta from going ahead with their separations.

Meta Says Staffing Decisions Came From People

The allegations were without merit and had no factual basis, a spokesperson for Meta said. The spokesperson said decisions on staffing and the way Meta was organised came from people, and not from AI.

The dispute comes at a time when the use of AI at work is raising broader questions about the law.

Workday is facing a separate high-profile case, with allegations that automated applicant screening disproportionately rejected people older than 40, as well as people with other protected characteristics. For the AI market, a bigger question is how much weight employers give a system’s results, even if people are formally making the final decision. People being involved may not solve the problem, if a score that is flawed still carries enough weight to determine who keeps a job.

Subscribe to our weekly news digest.

Join now and become a part of our fast-growing community.

Invalid Email Address
Thanks for Subscribing!
We'll be sending you our best soon!
Something went wrong, please try again later