Copilot reviewers speak out on grading AI-generated explicit content
Human contractors assigned to review Microsoft Copilot prompts and user-uploaded images are being exposed to sexually explicit and potentially illegal content, according to a report by 404 Media. The investigation cites Microsoft documents, contractor communications, and guidelines from Prolific, the company hiring reviewers for the task. These contractors are tasked with evaluating user interactions with Copilot to improve the AI assistant’s performance and safety features.
Reviewers report being inundated with explicit content requests, ranging from AI-generated sexual imagery to fake celebrity pornography. The volume and nature of this material raises serious concerns about contractor wellbeing and the psychological toll of regularly engaging with disturbing content.
While human review of AI model outputs is a standard industry practice for safety and quality assurance, this situation highlights labour concerns in content moderation roles. The apparent lack of adequate protections, support systems, or mental health resources for workers handling such material underscores broader issues within the content moderation sector. The practice also illustrates the tension between deploying large-scale AI tools and ensuring appropriate safeguards for the human workers responsible for oversight.