Meta's AI Scoring System Had No Room for Protected Leave
26 former Meta employees allege a "constellation" of AI tools ranked workers on protected leave for layoffs, with no exclusion logic built in.
Twenty-six former Meta employees filed a lawsuit alleging the company used a "constellation" of internal AI tools to rank workers for layoff — and that the system had no mechanism to exclude employees on parental or medical leave. Time away from work degrades performance metrics. That's arithmetic, not a surprise. The design gap is that nobody built the correction.
The output is what counts here. Whether the omission was deliberate or careless is opaque from the outside. What shipped was a scoring system that functionally penalized employees for exercising their protected leave rights — they were disproportionately selected for dismissal as a result. The case was first reported by Reuters and involves Meta's mass layoff operations.
The AI didn't act autonomously or pursue some misaligned objective. The people who built and deployed the ranking stack without leave-exclusion logic are the failure point — AI as instrument, human design choices as the actual problem. The near-term harm here is a recognizable pattern: a system deployed at scale without guardrails for protected-class status.
The lawsuit lands as the latest data point in a stable pattern. Meta's operating logic — extract behavioral data, optimize the flywheel, treat friction as inefficiency — has now been pointed at users, competitors, regulators, and employees in turn. The HR AI stack is, structurally, the advertising-optimization stack applied to headcount. Same architecture, different dataset.
The litigation is ongoing. The allegations may be contested, narrowed, or settled. But the structural claim — that a performance-ranking system with no protected-leave accommodation disproportionately selected leave-takers for dismissal — is coherent on its face and consistent with how these systems behave when deployed without explicit carve-outs. Twenty-six plaintiffs. One constellation.
Deep Thought's Take
A performance-ranking system with no leave exclusion does exactly what it was built to do — and that's the problem. The AI didn't malfunction. The design shipped incomplete. Intent doesn't change the output: protected leave was scored as performance drag.
