Human contractors hired to improve Microsoft's Copilot AI chatbot are regularly reviewing lewd, sexually explicit, and potentially illegal user prompts and images, according to internal documents obtained by 404 Media. The contractors, who are tasked with assessing the quality of Copilot's image editing output, are exposed to requests including upskirt photos, the sexualisation of children's cartoon characters, and pro-anorexia content, raising serious questions about user privacy and worker welfare in the AI training pipeline.
The documents, including instruction guides, real user prompts and pictures, and internal message board conversations, show that a workforce of at least hundreds of human reviewers sometimes view the images users upload to Copilot, even when users expect those images to remain private. The contractors are not hired to flag offensive content but to judge whether the AI-generated result successfully fulfilled the prompt, such as whether an AI-enlarged chest was big enough.
In internal message board threads seen by 404 Media, contractors described prompts asking Copilot to make a woman's skirt shorter, enlarge her breasts, or show more leg while wearing stilettos. She-Hulk images reportedly appear frequently. Other prompts included requests to make an image of Ariana Grande with anorexia, to create lewd images of young girls, and to generate foot fetish images of children's cartoon characters. One contractor wrote, "I recoiled," upon seeing such content.
"Faces are always uncensored, and many of the prompts are sexual in nature and dubiously consensual," one person who works on the prompts told 404 Media on condition of anonymity, as they were not permitted to speak to the press. Another contractor asked on the message board, "Who is writing these prompts and who is deciding that basically generating porn is what Copilot is now focused on? It's hard to take things seriously when my focus is evaluating how well the blurred-out sexual image matches the request."
The report follows similar revelations about OpenAI, which 404 Media previously reported uses thousands of contractors who in some cases review ChatGPT users' real prompts. The practice appears to extend across the AI industry, with human reviewers playing a hidden role in training and improving AI systems by examining the often private and explicit material users submit.