Microsoft's Copilot Reviewers See Your Uncensored Face. Opting Out Isn't an Option.
A 404 Media investigation found Microsoft routes some Copilot image edits, including the user's uncensored face, to outside contractors hired through Prolific. Microsoft's own privacy FAQ says there is no way to opt out.
By Mara Voss, Technology
· 3 min read · Updated
Key Takeaways
- •404 Media reported on September 29 that Microsoft routes some Copilot image-edit requests, including the uploaded photo's uncensored face, to outside contractors for review.
- •Contractors are hired through the platform Prolific and are paid to grade how well an edit matches the user's prompt, not to flag abusive or nonconsensual requests.
- •Microsoft's privacy FAQ states that an opt-out of human review is not available, even when a user disables AI model training.
- •In March 2024, Microsoft engineer Shane Jones told the FTC he had found more than 200 concerning images from Copilot Designer, the tool's predecessor.
- •Prolific did not respond to 404 Media's request for comment on how it trains or screens the reviewers handling this material.
Microsoft's Copilot image editor sends some photo-edit requests to outside contractors for human review. The face in the original photo is not blurred, cropped, or removed first. 404 Media reported the arrangement on September 29, citing contractors recruited through the research platform Prolific. They described what actually lands in their review queue: a real uploaded photo and the user's exact prompt, graded for how well Copilot followed instructions.
The short answer
Microsoft pays contractors, recruited through the platform Prolific, to review some Copilot image-edit requests. That review includes the user's original uploaded photo with the face left uncensored. Contractors described requests to shorten skirts, enlarge body parts, and sets of upskirt photos. Microsoft's own privacy FAQ says an opt-out of human review is not available. That is true even for users who turn off AI model training in their settings.
0
ways to opt out of Copilot's human photo review
Turning off the AI training setting in account preferences does not remove a photo from the review queue, per Microsoft's own privacy FAQ.
What reviewers are actually asked to do
In a typical assignment, a contractor sees the user's prompt, the original photo, and two edited versions Copilot produced. The job is to pick which edit follows the instructions better, keeps the rest of the image unchanged, and avoids visible distortion. Nothing in that task asks the contractor to flag content as abusive, nonconsensual, or illegal, according to contractors who described the workflow to 404 Media.
The content they described goes well past ordinary photo retouching. Contractors told 404 Media about requests to shorten skirts, enlarge body parts, and place real people in sexual positions. One reviewer said, "I just came across an image set that consisted of eight upskirt photos." Another described a set of edits applied to a real uploaded photo to show signs of anorexia.
Grading, not flagging
Contractors are paid to rate output quality, not safety. One reviewer asked: "Who is writing these prompts and who is deciding that basically generating porn is what Copilot is now focused on?" 404 Media said its reporting draws on interviews with contractors and internal documents it says it obtained; those documents have not been independently reviewed here.
Microsoft's answer, and what it leaves out
“Microsoft uses customer data as described in our terms of use, including to improve our products and enforce our code of conduct.”
That line does not say whether every Copilot image upload gets routed to a reviewer, or only some. It does not name how many contractors work the queue, or what training Prolific gives them for this material. 404 Media said Prolific did not respond to a request for comment.
This is not Copilot's first flagged safety gap
Microsoft has been here before, on a different part of the same feature. In March 2024, Microsoft principal software engineering lead Shane Jones wrote to then-FTC Chair Lina Khan. He had spent months testing Copilot Designer, the image generator that preceded today's editor. Jones said he found more than 200 examples of what he called concerning images, generated from ordinary prompts, including sexual and violent content. He raised the issue inside Microsoft first, by his account, and went to the FTC only after the company did not act.
Microsoft adjusted Copilot Designer's guardrails after that letter became public. The 2024 complaint was about what the software generated on its own. This one is about who gets to look at what a real person uploaded and typed, face included, with no setting that turns that off.
- Microsoft Copilot
- Prolific
- 404 Media
- AI content moderation
- data privacy
- Shane Jones
Sources
- 01Humans Are Reading Copilot Prompts, and They're Horrified, 404 Media404media.co
- 02Copilot users' photos reach human reviewers uncensored, 404 Media reports, The Next Webthenextweb.com
- 03Human Contractors Are Seeing All Your Horny, and Creepy, AI Prompts, Futurismfuturism.com
- 04Microsoft engineer begs FTC to stop Copilot's offensive image generator, Tom's Hardwaretomshardware.com
Corrections
No corrections have been made to this article.
About the reporter
Technology Reporter, Trestlewire
I spent seven years as a product manager at a mid-size SaaS company before I ever wrote a sentence for pay, which means I have sat through more roadmap reviews than most people would tolerate in a lifetime. I watched a scheduling feature get rebranded three times before it shipped, and I watched a launch date slide past four straight quarters while the slide deck stayed exactly the same. That is where the question I still ask every day came from: does this actually ship, or is it a demo.
Read full bio and all stories →