Report: Microsoft Copilot Contractors Can Fully View User-Uploaded Photos, AI Prompts, and More

The user data privacy protection mechanism for Microsoft's AI assistant Copilot is facing scrutiny. Internal documents disclosed by media show that contractors employed by Microsoft can fully view images uploaded by users, prompts entered, and AI-generated edited results when using Copilot, including photos containing clear facial information.

The primary responsibility of these contractors is not content moderation, but providing human ratings for AI generation quality. They evaluate AI output results from the perspective of ordinary users to help optimize model performance. This Reinforcement Learning from Human Feedback (RLHF) mechanism is relatively common in current large model training and iteration, but this incident reveals obvious vulnerabilities in the data anonymization and privacy protection aspects of this process.

This incident highlights the data privacy challenges faced by AI applications during actual deployment. When using AI assistants to process personal images or sensitive information, users often assume by default that the data is only used for machine processing, ignoring the existence of human review. For tech giants, how to strike a balance between improving model capabilities and protecting user privacy will become an urgent compliance issue to resolve.

This page provides an editorial summary based on publicly available information. It is not a republished article. Use the source link below for the original report.