In the 90s someone proved–mathematically–that invisible watermarks (e.g. hidden in metadata or in the pixel data itself) in photos would always be removable. I searched for it but I couldn’t find it but it should be obvious: Merely changing the format of an image is normally enough to destroy such invisible watermarks.

Basically, the paper I remember proved that in order for a watermark to survive a change in format/encoding it would need to be visible because the very nature of digital photo formats requires that they discard unnecessary information.

Also, I’d like to point out that it’s already illegal to remove watermarks (without permission) while simultaneously being trivial (usually) for AI tools like img2img to remove watermarks.