GoKawiilContractors grading Microsoft Copilot image edits see sexualized AI content, report says
According to 404 Media, human contractors hired to rate Microsoft Copilot's AI image-editing outputs are shown users' original photos, prompts and edited results, some of which include sexual, disturbing or potentially illegal content. Contractors said their job is limited to judging how well an edit matches the prompt across quality metrics, not flagging inappropriate material. Workers described seeing requests to sexualize women's bodies, upskirt images, pro-anorexia content and sexualized images involving children's cartoon characters, with faces left uncensored.
GoKawiil's interpretation of the reporting above, not reported fact.
The report suggests a gap in Microsoft's content moderation pipeline, where the people closest to potentially harmful outputs are instructed only to judge edit quality rather than report abuse, which could let problematic material persist unaddressed. This raises questions about how AI companies balance rapid model evaluation against safeguards for exploitative or illegal content, and about the psychological toll on contractors exposed to such material without a reporting mandate.
- Copilot's human reviewers see uploaded photos and AI edits but are tasked only with rating edit quality, not moderating content.
- Contractors reported exposure to sexualized, non-consensual-seeming, and potentially illegal imagery, including content involving children's cartoon characters.
- The findings, reported by 404 Media, highlight possible gaps in Microsoft's safety review process for its AI image-editing tool.
Source: techspot.com — Skye Jacobs, 2026-09-29
Published there as: “Report finds that Microsoft Copilot's human reviewers can see users' uploaded photos and sexualized AI edits”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.