Quick Answer
- A 404 Media report (Sept 28) says contractors rate Copilot image edits and can see users’ uploaded photos and prompts.
- Reviewers reportedly see faces uncensored.
- Microsoft says it uses customer data “as described in our terms of use”.
- Microsoft’s privacy FAQ says an opt-out of human review is not available.
- You can turn off model training, delete your chat history, and avoid uploading sensitive photos.
What did the report find?
On September 28, 2026, 404 Media published a story by Joseph Cox. Based on internal documents and contractor accounts, it says people hired to improve Copilot’s image editor can see what users upload. TechSpot, The Next Web, Malwarebytes, and Cybernews all covered the report.
According to these outlets, a reviewer sees:
- The user’s original prompt.
- The photo the user uploaded.
- Two edits that Copilot made.
The reviewer then picks which edit is better. They judge things like whether the edit followed the request, whether the rest of the image stayed the same, and whether the picture has visible flaws.
Can reviewers see faces?
Yes, according to the report. One contractor said, “Faces are always uncensored.” Cybernews and Malwarebytes repeat this. This claim comes from contractors quoted by 404 Media, not from Microsoft.
What kind of content did reviewers see?
The reports say many of the requests were sexual or disturbing, and some edits appeared to target real people without their consent. Contractors also described seeing harmful content, such as pro-anorexia material. Reviewers are reportedly paid to rate output quality, not to flag harmful requests, and their instructions tell them to trust their gut reaction.
We are not repeating the graphic details. The key point for you is simple: a human may see what you upload.
Who are the reviewers?
TechSpot names Prolific as one company that hires contractors for this kind of work. Prolific says it provides human feedback for AI tuning, safety checks, and benchmarks. It did not reply to 404 Media’s request for comment. 404 Media’s documents suggest at least hundreds of human reviewers sometimes see Copilot submissions.
What does Microsoft say?
A Microsoft spokesperson told 404 Media that the company “uses customer data as described in our terms of use,” including to improve products and enforce its code of conduct. The reports say Microsoft did not directly address the image-review work or the content reviewers saw.
Microsoft’s own Copilot privacy FAQ says:
- Some conversations are subject to automated and human review, for product improvement and digital safety.
- An opt-out of human review is not available.
- Conversations are stored for 18 months by default.
- Uploaded files are stored for up to 30 days.
- Images are de-identified for training, including removing metadata and blurring faces.
The FAQ does not say whether the image-rating work described by 404 Media is covered by the face-blurring step. Malwarebytes and All About Cookies both note that Microsoft has not cleared this up.
What can you opt out of?
| Control | Available? |
|---|---|
| Human review | No, per Microsoft’s FAQ |
| Model training on your chats | Yes, if you are signed in |
| Deleting chat history | Yes |
| Deleting Copilot memory | Yes |
Microsoft says that opting out of training covers past, present, and future conversations, and that changes take effect across its systems within 30 days. Turning off training does not appear to stop human review.
How to turn off training in Copilot
These steps come from Microsoft’s page for the older Copilot app. The updated Copilot app may look different, so check its Settings.
- Open copilot.com or the Copilot app and sign in.
- Click your profile icon.
- Open Privacy (in the Windows or Mac app: Settings, then Privacy).
- Turn off Training on conversation activity.
- Turn off Training on voice conversations.
How to delete Copilot data
- Delete conversations: you can remove single chats or your whole history from your conversation history.
- Delete memory: click your profile icon, open Memory, and choose Delete all Memory. This does not delete your chat history.
All About Cookies points out that deleting data will not undo a review that already happened, but it can shorten how long your data stays on Microsoft’s servers.
How to stay safe when you use AI image tools
- Do not upload private photos. Upload only what you would be fine with a stranger seeing.
- Do not upload photos of other people, especially children, without their consent.
- Crop out faces when the edit does not need them.
- Remove personal data from the file. On Windows, right-click the photo, choose Properties, then Details, then “Remove Properties and Personal Information”.
- Check the privacy settings of every AI tool before you use it.
Is this only a Microsoft problem?
No. Cybernews notes that related reporting says OpenAI has also hired people to read ChatGPT prompts, which suggests human review is an industry-wide practice. If you want to compare, read our guide on whether ChatGPT is safe. For another example of AI tools reaching into your private data, see our report on macOS and AI agents.
Why this matters
Many people assume they are talking to a machine. This report shows a human may sometimes read your prompts and view your photos. The gap between what users expect and what actually happens is the real issue, and the only protection today is to be careful with what you upload.
FAQs
Can Microsoft employees see my Copilot photos?
The report says outside contractors who rate Copilot’s image edits can see uploaded photos and prompts. Microsoft’s FAQ also confirms that some conversations are reviewed by humans.
Can I opt out of human review in Copilot?
Microsoft’s FAQ says an opt-out of human review is not available.
Can I stop Copilot from training on my chats?
Yes, if you are signed in. Turn off the training toggles in Privacy settings.
Does deleting my Copilot history help?
It removes your conversations and limits how long your data is stored, but it cannot undo a review that already happened.
Are faces blurred before reviewers see them?
Microsoft’s FAQ says faces are blurred for training. A contractor told 404 Media that faces are always uncensored in the review tasks. Microsoft has not explained the difference.
Is Copilot safe to use?
For general questions it is fine. Avoid uploading sensitive photos, documents, or images of other people.
Is this the same in other AI apps?
Reports say other companies also use human reviewers. Read each app’s privacy policy to see the details.
