Researchers Tested AI Watermarks—and Broke All of Them

In a new paper out this week, a group from the University of Maryland found it's too easy to evade current methods of watermarking—and even to add fake watermarks to real images.