The Meta Case Exposing a Dangerous Blind Spot in AI Performance Reviews

Earlier in the summer 26 Meta employees alleged that internal AI systems helped score, rank and select workers during layoffs affecting around 8,000 jobs. According to the complaint, employees taking protected leave appeared disproportionately on the termination list because the systems treated reduced digital activity as weaker performance.

One plaintiff, a scientist on approved pregnancy leave, reportedly learned about her layoff two days before giving birth. Another employee said time away for an injury contributed to a lower rating. A manager on medical leave was informed that his role would end only 16 days into his absence.

Meta disputes the allegations, noting, “These claims lack merit and are not based on facts. Workforce management and organizational decisions were and are made by people, not AI.” (The Guardian, 2026)

Activity Data Can Miss the Reason Behind It

Workplace AI is often sold as a more consistent way to assess performance. Every employee enters the same system, the same metrics are applied, and the same scoring process produces a ranking. Consistency at the level of measurement, however, says little about fairness.

An employee on medical leave will naturally generate fewer keystrokes, messages and completed tasks than someone working full-time. A worker with a disability may use different hours, tools or workflows. Once activity becomes a proxy for value, protected circumstances can begin to resemble underperformance.

Digital monitoring can show when someone moved a mouse or opened a browser. It cannot easily explain the quality of a decision, the complexity of a task or the reason a worker temporarily disappeared from the dataset.

Meta’s case therefore raises a larger problem: workplace systems are becoming highly capable at recording behavior while remaining far less capable of interpreting human context.

Human Oversight Does Not Erase Algorithmic Influence

Employers frequently point to human approval as a safeguard. Yet a manager reviewing an AI-generated ranking still begins with the assumptions built into that ranking.

A person may make the final decision while relying on scores shaped by incomplete data. Responsibility then becomes fragmented. The model generated the assessment, managers followed its output and executives approved the process.

California regulators have already moved to address that gap. Employment rules now clarify that automated decision systems can violate discrimination law when they contribute to decisions affecting protected groups. Employers must also retain relevant automated-decision records for at least four years, making later audits more feasible (Civil Rights Department, State of California, 2026).

Meta’s internal monitoring program reportedly triggered a petition signed by more than 1,600 employees before Mark Zuckerberg paused it in June (The Guardian, 2026). The backlash reflected a wider concern: companies are collecting increasingly detailed data about workers before establishing clear limits on how it should influence careers.

AI can rank thousands of employees in seconds. Fair judgment requires something slower and more demanding: understanding what the data leaves out.

The Meta lawsuit may ultimately turn on specific evidence. Even when an algorithm converts human circumstances into a score, employers still carry responsibility for asking whether the score reflects reality.