A 404 Media report says contractors hired by third-party companies are reviewing some Microsoft Copilot users’ uploaded photos, image-editing requests and generated results. Reviewers are shown an original image, the user’s instruction and two possible edits, then asked to assess technical quality, including whether the request was followed, whether details were preserved and whether the result looks coherent. Faces in the source images are reportedly visible.
The material described includes sexualised and apparently non-consensual edits, such as shortened skirts, enlarged body parts and sexual positions. Contractors told 404 Media they had seen upskirt images, foot-fetish images involving children’s cartoon characters, an image request depicting Ariana Grande suffering from anorexia, and animal-sacrifice imagery. They said their role was to judge image quality rather than identify illegal or unsafe content. The report does not establish how widespread the reviews are.
Microsoft’s privacy FAQ for the older consumer Copilot says conversations are retained for 18 months by default and that uploaded images or files may be used for AI training. Eligible users can opt out of training, but the FAQ says an opt-out from certain human reviews is unavailable. Microsoft did not tell 404 Media whether users can prevent uploads from being included in the image-rating process.
Microsoft says its updated consumer Copilot app, available since 18 August, does not use prompts, responses or uploaded file contents to train foundation models; it has not clarified whether that excludes human review.