Researchers at the University of Waterloo have developed UnMarker, a tool capable of removing watermarks from AI-generated images in minutes. Despite investments in digital watermarking to secure image provenance, lead researchers Kassis and Hengartner assert that such watermarks can be deleted regardless of their encoding. This raises concerns over the effectiveness of current AI safety measures. Kassis highlights the need to assess the true protective abilities of watermarking technology given the rising issue of harmful AI content, including scams and exploitative imagery.
Kassis stated, "It's no secret that we're surrounded by AI wherever we go. And although it has many benefits, it also has a dark side unfortunately."
Kassis emphasized the need to question, "Is it worth the hype? Can it really protect us or are we still vulnerable?"
Digital watermarks intended to secure image provenance can be removed, challenging the effectiveness of existing AI safety measures like Google’s SynthID and Meta’s Stable Signature.
The software tool UnMarker can erase watermarks in AI-generated images using a 40 GB Nvidia A100 GPU, operating offline and taking only a few minutes.
Collection
[
|
...
]