API
< Back to all clusters
[TECHNOLOGY] · United States · 12 sources

started · updated

Microsoft Copilot contractors review uncensored user photos and prompts

Internal documents obtained by 404 Media reveal that Microsoft employs hundreds of third-party human contractors to review user data processed through its Copilot AI system. These contractors evaluate the accuracy of AI-generated image edits by comparing original user-uploaded photographs with the resulting AI outputs.

Reports indicate that these reviewers often see uncensored images, including those with identifiable faces, alongside the original text prompts. The evaluation process is designed to measure how well the algorithm follows specific instructions rather than serving as a safety moderation service. Consequently, contractors have reported encountering highly disturbing and sensitive content, including sexually explicit requests, non-consensual image modifications, and prompts involving children or self-harm.

Microsoft has defended the practice, stating that it uses customer data in accordance with its terms of use to improve products and enforce conduct standards. While the company offers harmful content protection, certain settings can be disabled by organization administrators. One of the companies involved in providing human evaluation for these tasks is Prolific.

Entities

404 Media · Copilot · Microsoft · OpenAI · Prolific

Claims

What the coverage asserts, and how many sources carry each claim.