Analysis-Meta employees' lawsuit shows that if AI fires you, proving it is the hard part

By Daniel Wiessner

July 22 (Reuters) - A novel lawsuit claiming that Meta Platforms relied on discriminatory AI tools to select employees for layoffs highlights the problems workers face in suing employers over the new technology, including proving how it was actually used. 

The case helps illustrate why a widely predicted wave of employment lawsuits over AI use has yet to arrive. Legal experts say workers often have little understanding of how AI systems are used in the workplace and many have also signed away their right to sue in court, agreeing instead to resolve workplace disputes through a private process called arbitration that can keep such claims from ever being tested publicly.

In a ruling last week declining to block Meta from finalizing the terminations of 26 people who sued, U.S. District Judge William Orrick identified a fundamental obstacle for plaintiffs who allege that AI discriminated against them: "they were not in the rooms where it happened."

That means workers like the Meta employees, who claim they were targeted for layoffs because they have disabilities or took medical or family leave, often cannot muster the evidence of wrongdoing necessary to quickly secure a win in court.

And they face another obstacle: Like a majority of U.S. workers, the plaintiffs are bound by an arbitration agreement, meaning they cannot band together in a class action, put their case before a jury, or push for a multimillion-dollar settlement in open court.

ARBITRATION AGREEMENTS BLOCK LAWSUITS

Companies generally prefer arbitration, which they say is ⁠a faster, cheaper alternative to court, while worker advocates say it often favors employers and discourages workers from bringing claims. The arbitration process is also confidential, so it can shield unfavorable evidence unearthed in an individual case from wider disclosure.    

"Even if you establish that a particular system would produce discriminatory outcomes left and right, you have no way of sharing that information with other employees," said Christine Webber, co-chair of the civil rights and employment practice at plaintiffs' firm Cohen Milstein Sellers & Toll. Webber's firm is not involved in the Meta case. 

Webber and other plaintiffs' lawyers said those hurdles explain the lack of high-profile court cases involving employers' use of AI even as it becomes routine, and why even the lawsuit against Meta seeking only temporary relief is unusual. 

One of the few cases to emerge over companies' workplace use of AI tools involves Workday, which is facing claims that its popular HR management software unlawfully filtered out applicants for jobs at other companies based on race, age and disability. Arbitration is not an issue in that case because Workday does not have agreements with its customers' job applicants. Workday denies the allegations.

PLAINTIFFS SEEK INJUNCTION

The agreements signed by the Meta workers contain a common, narrow exception for seeking a court order that temporarily blocks one side from taking some irreversible action. But that exception is typically invoked in cases involving the alleged theft of trade secrets or the solicitation of clients or employees, ​and not layoffs of at-will workers.

Orrick denied the plaintiffs a temporary restraining order that would have stopped Meta from completing the layoffs. He must still decide whether to issue a preliminary injunction, a temporary but longer-lasting order that would put the workers back in their jobs until their individual arbitration cases are resolved. He said he could change his mind and grant the injunction if the plaintiffs come up with evidence "regarding whether and how AI was used in an improper manner."

A hearing is scheduled for August 24, and the losing side can appeal Orrick's decision. 

The plaintiffs claim that in selecting jobs to cut, Meta consulted AI tools that tracked productivity and AI token usage (a measure of how much workers use AI tools), disadvantaging people who missed work because of medical conditions or to care for family members.

They allege that Meta used a number of internal AI-assisted systems including a large language model assistant known as "Metamate," an employee-trained "second brain" that tracked workers' communications and documents, ​and a productivity score drawn from scanning keystrokes, screen content, emails and browser history, according to the lawsuit.

Meta said in court filings and statements last week in response to the lawsuit that humans made all of the decisions concerning nearly 8,000 layoffs announced earlier this year and has denied treating AI usage as a basis for identifying workers to terminate or to conduct performance reviews. A Meta spokesman said on Tuesday that the company had no further comment. 

Orrick said in his decision that he was bound to take Meta at its word since the plaintiffs could not present any evidence to rebut those claims.

The plaintiffs' lawyers in a joint statement last week acknowledged the hurdles they face in gathering evidence, even calling on current and former Meta employees to contact them with knowledge of how AI was used in the selection process.       

"Meta holds virtually all the relevant information," they said.

(Reporting by Daniel Wiessner in Albany, New York; Editing by Alexia Garamfalvi and Matthew Lewis)