On January 18, 2023, TIME journalist Billy Perrigo published an investigation revealing the human cost behind ChatGPT’s safety features. To build a tool that could detect and filter toxic content, OpenAI had, starting in November 2021, contracted the outsourcing firm Sama to have workers in Kenya label tens of thousands of text snippets describing graphic violence, child sexual abuse, bestiality, murder, suicide, torture, self-harm, and incest. The labeled examples taught a classifier to recognize toxicity so that ChatGPT could screen it out before it reached users.
The workers who did this labeling were paid a take-home wage of roughly 1.32 to 2 dollars an hour, depending on seniority and performance. All four employees TIME interviewed described being mentally scarred by exposure to the material. They were nominally entitled to sessions with wellness counselors, but said those were rare and unhelpful amid pressure to hit productivity targets. The trauma was severe enough that Sama canceled all of its work for OpenAI in February 2022, eight months earlier than the contract had planned.
The story became a defining example of the hidden labor that makes “automated” AI possible, the very phenomenon Mary Gray and Siddharth Suri had named ghost work. It exposed an uncomfortable economic truth: the safety and polish of a flagship AI product rested on low-paid, psychologically hazardous human work outsourced to the Global South. For business and policy readers, it is a lasting reminder that AI supply chains include people, and that the cost of cleaning up a model is often paid by the most vulnerable workers in it.