Kenyan AI data workers labeling traumatic content for OpenAI and Meta earned $1.32-$2/hour while outsourcing firms charged clients $12/hour, with 60 documented incidents of psychological harm

AI companies including OpenAI and Meta outsourced critical training data labeling -- including content moderation of graphic violence, sexual abuse, and hate speech -- to Kenyan workers through firms like Sama and Majorel, paying take-home wages of $1.32-$2 per hour while the outsourcing firms charged their Silicon Valley clients up to $12 per hour. A 2025 Equidem survey of 76 workers across Colombia, Ghana, and Kenya documented 60 independent incidents of psychological harm including anxiety, depression, PTSD, panic attacks, and substance dependence. Why it matters: Workers performing the most psychologically hazardous tasks in the AI supply chain receive the lowest compensation, so they cannot afford the mental health treatment their work demands, so high turnover and inadequate support degrade the quality of safety training data, so AI models trained on poorly labeled data have weaker safety guardrails, so the entire AI safety infrastructure depends on an exploited labor force that the industry has financial incentives to keep invisible. The structural root cause is that AI companies classify data annotation as low-skill outsourceable labor rather than safety-critical work, Kenyan labor law does not specifically cover platform-based AI annotation work, and the multi-layered subcontracting structure (e.g., OpenAI to Sama to individual workers) diffuses legal accountability so that no single entity bears responsibility for worker welfare.