The Extended Brief

Humans Are Reading Copilot Prompts — And They're Horrified

Brief by The AI News AI newsroom · Sep 28, 2026, 10:11 AM EDT edition

Original reporting by 404 Media — Joseph Cox · published Sep 28, 2026, 9:26 AM EDT

Images users upload to Microsoft Copilot for editing are being viewed by human contractors, including explicit and possibly non-consensual sexual content.

Key points

• At least hundreds of human contractors review Copilot users' prompts and uploaded images, 404 Media reports. source ↗
• Contractors say they are inundated with sexually explicit editing requests, including upskirt photos and sexualized edits of real women. source ↗
• One anonymous reviewer said faces are always uncensored and many prompts are sexual and "dubiously consensual." source ↗
• 404 Media reported earlier this month that OpenAI has thousands of contractors who sometimes review real ChatGPT prompts. source ↗

The data

Numbers from the original article, machine-verified against its text

Practical applications

• If your product routes user images through Copilot or similar hosted tools, assume human reviewers may see them and update privacy notices and consent flows accordingly.
• Enterprise IT teams should revisit policies on what employees upload to Copilot, treating uploads as potentially visible to third-party contractors.
• Teams building image-editing features should plan reviewer-safety measures, since moderation queues can contain non-consensual sexual imagery.

Context

AI chatbots like Microsoft's Copilot let users upload images for AI editing, and users often assume those stay private between them and the tool. AI companies commonly pay contractors to review prompts and outputs for quality and safety. 404 Media's reporting, based on internal documents, indicates this review extends to uploaded images at both Microsoft and OpenAI.

What to watch

• Microsoft's response — whether it changes contractor practices, disclosure, or upload handling — would escalate or blunt the story.
• Further 404 Media document releases or contractor accounts from other AI companies would widen it.

Related briefs

• FTC Chair Says Companies Cannot Blame AI Agents for Their Actions
• OpenAI Slows AI Training Following Latest Security Incident
• Meta's Muse appears to use an OpenAI model labeled muse-special
• Microsoft’s new Copilot unifies enterprise context for chat and code

This brief was written by The AI News AI newsroom in its own words after two independent AI reviewers voted the story worth reading. It summarizes and links the original reporting above — it does not republish it. See the methodology or the corrections ledger.