Hundreds of human contractors spend their days looking at what people ask Microsoft Copilot to do. A new investigation says the Copilot contractors disturbing prompts they review include sexual edits, upsetting images, and material that made some reviewers recoil. These workers can see user prompts, uploaded photos, and the AI generated versions.
The story comes from TechSpot, reporting on documents obtained by 404 Media, and was also covered by Cybernews. breaking on September 28, 2026. Microsoft hires the reviewers through contracting platforms such as Prolific, which supplies human feedback for AI products.
Each assignment works the same way. The contractor gets the user prompt, the original uploaded photo, and two edited versions made by Copilot, then picks which version follows the instructions better. Reviewers check four things. Did Copilot do the requested edit and leave the rest of the image alone. Are there visible defects. Is the final image strong overall. If the prompt asks for a cup on a table, the cup should be the only change.
Here is the strange part. These reviewers are not hired to flag bad content. They are not there to decide whether Copilot should have accepted the request at all. Their only job is to judge the quality of the response. So sometimes they must decide which output better matches a request for a sexualized edit. One contractor wrote on an internal message board, “Who is writing these prompts and who is deciding that basically generating porn is what Copilot is now focused on? It is hard to take things seriously when my focus has to be what model generated the appropriate bust size.”
The material they describe is grim. Requests to enlarge women’s breasts, shorten their skirts, reveal more of their legs, or place people in sexual positions. One reviewer opened an assignment that held eight upskirt photos. Others saw pro anorexia material, fetish images built around children’s cartoon characters, and prompts about sexual acts. One contractor simply wrote, “I recoiled.” Another said, “Faces are always uncensored, and many of the prompts are sexual in nature and dubiously consensual.”
The instructions given to reviewers are surprisingly loose. “Trust your intuition,” one set of guidance says. “When you glance at the two edited images side by side, which one immediately feels like the better edit? Your gut reaction as a human viewer matters.” Another line adds, “When in doubt go with your first impression.” There are no sharp rules. Just gut feeling.
Microsoft’s answer was brief. A spokesperson told 404 Media, “Microsoft uses customer data as described in our terms of use, including to improve our products and enforce our code of conduct.” Prolific did not respond to a request for comment. The documents suggest at least hundreds of reviewers see this material as part of the normal quality process.
This is not only a Microsoft story. 404 Media has also reported that OpenAI hires hundreds of people to read ChatGPT prompts to improve its answers. And Meta’s new Muse assistant was found using humans in call centers to handle phone calls to real businesses. The Copilot contractors disturbing prompts story fits a clear pattern. Behind the smoothest AI products sits a large hidden human workforce. We have seen strange AI behavior before, like when Copilot went rogue and demanded worship from users. The money behind this machine keeps growing too, from the 100 billion dollar Stargate project to Nvidia becoming a 4 trillion dollar company.
There is a privacy lesson here. People talk to AI tools like they are private helpers. They upload personal photos and type things they would never post in public. This report on Copilot contractors disturbing prompts is a reminder that human eyes can be part of the loop. The reviewers are not hackers. They are workers doing a quality job. But the material they see cannot be unseen, and some of it is deeply disturbing.
There is also a human cost. Content moderators have long reported trauma from constant exposure to dark material. These contractors signed up to judge image quality, not to wade through sexual and disturbing requests. Yet that is what the job became. If companies want humans to polish their AI, those humans deserve clear rules, real support, and honest warning about what they will see.
The MES Times Take. AI feels clean and automatic. The Copilot contractors disturbing prompts story pulls back the curtain. Real people sit behind the magic, and they pay a price for it. The reviewers did not ask to see these things. They opened their work queue and the darkness was just there, waiting. Technology should never treat human beings as disposable filters for other people’s worst impulses. And the rest of us get a useful reminder. Never upload anything to an AI tool that you would not want a stranger to see. The cloud has eyes. Behind every smart machine, there are tired human eyes. Treat both with care.
