A technical investigation published this week on hackernews has revealed that Microsoft's MS Paint and Photos applications — both bundled with standard Windows installations — are embedding invisible watermarks containing globally unique identifiers (GUIDs) into image files saved on users' own local machines. The finding, documented through hands-on reverse engineering at xusheng.dev, shows this behavior applies even when no internet connection or cloud service is involved and regardless of whether the image was AI-generated or a conventional photograph edited locally.
The watermark is not stored in standard EXIF or IPTC metadata fields, where it would be visible to any ordinary metadata viewer. Instead it is embedded in a manner that survives typical "strip metadata" workflows, according to the researcher's analysis. The GUID structure used is consistent with identifiers that can be tied back to a specific Windows installation or device, meaning two files saved on two different machines would carry different identifiers even if the image content were pixel-for-pixel identical.
Microsoft has not, as of publication, issued a public statement explaining the purpose of the watermarks, the data-retention policy around any server-side logging of those identifiers, or whether the behavior is disclosed in its Windows license terms or privacy documentation. The scope is notable: MS Paint and Photos are not optional downloads — they ship with Windows 11 by default and are among the most widely installed image applications on the planet, covering a user base that numbers in the hundreds of millions.
The timing lands in the middle of an active international policy debate. The EU AI Act's provisions on synthetic-content labeling and the US executive activity around AI watermarking standards have focused almost entirely on watermarks applied by AI generators at the point of creation. This disclosure demonstrates that watermarking is already happening at the point of saving and editing, silently, by the operating system itself, and on content that has nothing to do with generative AI — a distinction that existing regulatory frameworks and most public commentary have not addressed.
For preppers and self-reliant households, the detail that matters most is one that general tech coverage is likely to underemphasize: this watermarking happens to locally stored files with no apparent network trigger, which means the identifier is baked in at rest and will travel with the file any time it is shared, submitted as documentation, uploaded to a legal proceeding, or handed to an insurance adjuster. Households that photograph property inventories, damage documentation after a disaster, or sensitive personal records using Windows-native tools are generating files that carry a machine-linked fingerprint they almost certainly do not know exists. That fingerprint could, in principle, be used to establish which device produced a given image — a capability with legitimate forensic uses but also with obvious implications for anyone whose documented property records, medical images, or communications might later be subject to scrutiny they did not anticipate. Our overview of offline-capable image tools is one place to find editors that do not exhibit this behavior.
