Overview
- The reporting, published Monday, September 28, 2026, relied on internal instruction guides, message‑board posts, and example tasks to show at least hundreds of contractors review Copilot users’ image‑editing prompts, uploaded photos, and the AI’s two candidate edits.
- Reviewers are hired through third‑party vendors and paid to judge which edit best follows a prompt by using “human intuition,” not to act as content moderators or make legal assessments of images.
- Contractors say they see uncensored faces and frequent sexual or exploitative content, including upskirt photos, sexualized images of young girls or cartoon characters, pro‑anorexia requests, and other disturbing material that reviewers reported was traumatic.
- Microsoft told reporters it uses customer data as described in its terms of use, and public Copilot privacy materials note stored conversations and that some uploaded images may be used to improve products while certain human‑review opt‑outs may not be available.
- The disclosures raise immediate questions about user privacy, vendor oversight, and protections for low‑paid reviewers, and they could lead to demands for clearer opt‑outs, stronger vendor rules, and regulatory scrutiny if companies do not change practices.