Graded by algorithms, monitored right down to the keyboard, then fired: 26 Meta employees tell the story of a cart driven by AI. A method which, in Europe, would already have a price: up to 4% of turnover.
Meta’s spring cleaning ends in court. Monday evening, 26 employees and former employees filed a complaint in federal court in Oakland, California. According to them, the list of 8,000 positions deleted in May (10% of the workforce) was not drawn up by managers, but by a battery of artificial intelligence systems. Meta denies and assures that these decisions were made “by humans, not by AI”.
How would the algorithm have chosen the starters?
The complaint, 71 pages long, describes a « constellation » homemade tools. We find there Metamatethe internal assistant powered by large language models, and agents “second brain” trained on the documents and exchanges of each employee. Added to this are readings from keystroke monitoring software deployed in the spring, dashboards measuring AI token consumption and algorithm-assisted performance rankings.
The heart of the accusation lies in the mechanics of these scores: built on measured productivity and recorded activity, they function as a continuous monitoring where absence is worth zero, including sick leave. Employees on sick leave, on parental leave or with a disability would therefore, by construction, never have had the means to catch up with their colleagues, and would have found themselves over-represented on the list. The 26 plaintiffs (anonymous, from six states and Washington) are requesting a freeze on the layoffs, effective July 22, while they take their cases to arbitration.
And in Europe, would this be legal?
The short answer lies in an article of law: since 2018, the GDPR prohibits a decision producing legal effects (a dismissal obviously ticks the box) from being based exclusively on automated processing. Real human intervention is required, not a simple click of validation on the machine’s output, otherwise the cost could rise to 20 million euros or 4% of global turnover.
The AI Act was to add a second lock: employee management systems, from recruitment to dismissal, are classified as high risk, with bias tests, human supervision and employee information. These obligations were to apply on August 2, except that Brussels has just postponed the entire aspect to December 2027 (the sense of timing, among legislators, remains an inexact science). The first major trial of algorithmic dismissal therefore breaks out the same week that Europe postpones its safeguards.
👉🏻 Follow tech news in real time: add 01net to your sources on Google, and subscribe to our WhatsApp channel.
Source :
Reuters
