The fact that the watermark detection algorithm doesn't seem to require a tremendous amount of processing power unlike most image processing/detection algorithms - apparently it's present in scanners and photocopiers too - implies that it could also be embedded in camera firmware, making it very possible to produce cameras that refuse to take pictures of certain things... I find that more unsettling since scanners/photocopiers are somewhat more specialist equipment today, whereas practically everyone has at least one camera. Imagine not being able to take a picture of something crucial like a crime in progress because there happened to be something watermarked present in the scene. That's why I don't believe in leaving these types of moral decisions to the machine.

I think it's only the legal issues that keep people from RE'ing the algorithm and generating obviously-non-banknote images which get detected, in a similar spirit to this:

http://news.cornell.edu/stories/2015/03/images-fool-computer...

As the watermark detection algorithm is not that resource-intensive, I would guess that a false positive is even easier to produce.