A security researcher has reported that Microsoft embeds hidden identifiers into AI-generated images produced through Copilot in Paint and the Photos app, according to a report by Neowin. These watermarks are invisible to the naked eye but are baked directly into the image data itself, meaning every picture you generate with these tools may carry a fingerprint you can’t see.
Here is what we know about the discovery and what it could mean for anyone who uses Microsoft’s built-in AI image tools.
What the researcher found
The researcher says that images created using Copilot’s image generation features in both Paint and the Photos app contain embedded watermarks. These are not the visible stamps or labels you might expect burned into a corner of a photo — they are hidden data woven into the pixels of the image.
The identifiers appear in the output of Copilot when it generates images inside these two apps. The report focuses on confirming that the data is present and invisible, rather than on a detailed technical breakdown of the encoding format. Microsoft has not publicly commented on the specific findings as of the report.
For context, Copilot in Paint launched as part of Microsoft’s broader push to fold AI generation into everyday apps, letting you create and edit images from text prompts without leaving the familiar Paint interface. The Photos app has seen similar AI features added over time, which is why the watermark reportedly appears in both places.
How hidden AI image watermarks work
Watermarking AI-generated content has become a common practice across the industry, driven largely by concerns over deepfakes and AI-generated misinformation. Companies including Microsoft, Google, and others have explored ways to tag machine-made images so they can be identified later, even after being shared or edited.
Unlike a visible logo, hidden watermarks store information in the subtle variations between pixels. This makes them nearly impossible to see with the naked eye, yet they can sometimes be detected by software that knows how to look for them. The goal is usually to trace the origin of an image — which tool generated it, and sometimes when.
This technique is often used to help platforms and researchers distinguish human-created content from AI-generated content, giving them a way to flag machine-made images at scale.
Privacy and detection concerns
The discovery raises questions about privacy and how your images might be tracked. If these watermarks can be detected by third-party software, then images you export from Paint or Photos could carry a hidden marker that reveals their AI origin.
Some concerns revolve around social media platforms. There have been reports that platforms like Meta and TikTok use watermark detection to label or demote AI-generated content. If Microsoft’s watermarks are detectable by these systems, images you download could be flagged automatically when you upload them.
There is also a broader debate about whether users should be told when their images carry hidden identifiers. Transparency advocates argue that people should know if the files they create contain embedded tracking data, especially when that data could be picked up by external services beyond their control.
What This Means for You
If you use Copilot to generate images in Paint or the Photos app, your output may now contain a hidden watermark. This is unlikely to affect casual use — the images will look completely normal, and there is no indication the watermark degrades visual quality.
However, if you plan to share AI-generated images on social media, be aware they might be automatically detected and labeled as AI-made. This is not unique to Microsoft; many AI image generators now embed similar identifiers as the industry responds to deepfake concerns.
What to Do
If you want to avoid carrying a hidden watermark, your best option is to generate images with a tool that does not embed one, or to simply be aware that Microsoft’s Copilot-powered apps may add this data to your files.
You can also check the provenance of AI images you receive using detection tools, though these are not always reliable and can miss or misidentify watermarks. For most users, the practical takeaway is simply to know that AI-generated images from Microsoft’s apps may carry invisible identifiers.
The bigger picture
Microsoft has been steadily integrating Copilot AI features into everyday apps, from Paint to Photos to Word. Adding watermarking to AI-generated images aligns with a wider industry push to make AI content identifiable, even as it sparks ongoing debate about transparency and user consent.
As AI image generation becomes more common, expect more tools to adopt some form of content tagging. The conversation about how much users should control — and be told — about these identifiers is likely to continue.
Source: Neowin
Over to you: Will you keep using Copilot in Paint to generate images, or switch to a tool that doesn’t embed hidden watermarks?
