OpenAI, the artificial intelligence company behind ChatGPT, paid workers in Kenya less than $2 an hour to review graphic and disturbing content as part of its safety training, according to a Time magazine investigation. The workers, employed by the San Francisco-based data annotation firm Sama, were tasked with reading and categorizing text and images that included child sexual abuse, bestiality, murder, suicide, and torture, often described in graphic detail.
The arrangement, which began in November 2021, was meant to help OpenAI build tools to filter harmful content from its AI systems. But the toll on the workers was severe. One moderator told Time that reading such material felt like "torture," describing a particularly disturbing story about a man having sex with a dog in front of a child. "You will read a number of statements like that all through the week. By the time it gets to Friday, you are disturbed from thinking through that picture," the worker said.
Despite the emotional weight of the work, pay was minimal. Sama received $12.50 an hour from OpenAI, but after taxes and Sama's cut, workers took home roughly $1.30 to $1.50 an hour, potentially reaching $2 if they met performance metrics like accuracy and speed. Nairobi, where the workers were based, has no minimum wage requirement, but the disparity between the company's valuation—reportedly $29 billion in a deal with Microsoft—and the wages paid to these workers has drawn sharp criticism.
Sama, which also recently ended a long-term contract with Facebook, offered "wellness counseling" to support employees, but workers said those sessions were rare and often replaced by group meetings instead of the promised one-on-one therapy. Sama disputes this, maintaining that counselors were always available.
The relationship between OpenAI and Sama deteriorated further in February 2022, when OpenAI hired Sama to help with an image-generation project, likely related to DALL-E 2. Sama employees were asked to collect graphic images to train the model against visual content. Some of the material fell into a category known as C4, which includes imagery that is illegal under US law. When Sama discovered the issue, it immediately ended the image classification pilot and gave notice that it would cancel all remaining projects with OpenAI.
A Sama spokesperson told Time that the individuals working with the client did not vet the request through proper channels, and after a review, those individuals were terminated and new sales vetting policies were put in place. OpenAI, for its part, said it never explicitly asked for C4 content. "This content is not needed as an input to our pretraining filters and we instruct our employees to actively avoid it," an OpenAI spokesperson said. "As soon as Sama told us they had attempted to collect content in this category, we clarified that there had been a miscommunication and that we didn't want that content."
OpenAI also defended its broader mission, stating, "Our mission is to ensure artificial general intelligence benefits all of humanity, and we work hard to build safe and useful AI systems that limit bias and harmful content." The company maintains that classifying and filtering harmful content is a necessary step in minimizing violent and sexual material in training data.
However, the report raises questions about the ethics of outsourcing such psychologically demanding work to low-wage workers in the Global South. While content moderation is essential for the internet to function, the industry's reliance on underpaid labor in developing countries remains a persistent issue. As AI companies grow increasingly wealthy, the treatment of the workers who make their safety systems possible is likely to face continued scrutiny.