AI Watermarking Goes Quiet and Mandatory

Microsoft Paint and Photos are now adding both a visible and an invisible watermark to AI-generated content. The visible one can be turned off. The invisible one, tied to a GUID, cannot, and it happens without any user notification. The thread on this story surfaced both alarm and grudging acknowledgment that this might actually be useful infrastructure if enough platforms adopt it.

The pattern here: watermarking was widely dismissed as unworkable because it could be stripped. The GUID approach embedded at the software level, not the file level, is harder to strip without degrading the image. Several commenters noted that if cameras did the same with cryptographic signing, you could prove a photo was not AI-generated, which inverts the current assumption that provenance is unverifiable.

The counterpoint from the thread is real: this only works if adoption is broad enough to matter, and right now it is just Microsoft. If Adobe, Apple, and Google do not follow, the watermarks become meaningless signal buried in noise. But the direction of travel is clear.