Please create an account to participate in the Slashdot moderation system

 



Forgot your password?
typodupeerror
×
AI

Researchers Say Current AI Watermarks Are Trivial To Remove 9

Researchers from the University of Maryland (UMD) were able to easily evade the current methods of AI watermarking during testing and found it even easier to add fake emblems to images that weren't generated by AI. "But beyond testing how easy it is to evade watermarks, one UMD team notably developed a watermark that is near impossible to remove from content without completely compromising the intellectual property," reports Engadget. "This application makes it possible to detect when products are stolen." From the report: In a similar collaborative research effort (PDF) between the University of California, Santa Barbara and Carnegie Mellon University, researchers found that through simulated attacks, watermarks were easily removable. The paper discerns that there are two distinct methods for eliminating watermarks through these attacks: destructive and constructive approaches. When it comes to destructive attacks, the bad actors can treat watermarks like it's a part of the image. Tweaking things like the brightness, contrast or using JPEG compression, or even simply rotating an image can remove a watermark. However, the catch here is that while these methods do get rid of the watermark, they also mess with the image quality, making it noticeably worse. In a constructive attack, watermark removal is a bit more sensitive and uses techniques like the good old Gaussian blur.

Although watermarking AI-generated content needs to improve before it can successfully navigate simulated tests similar to those featured in these research studies, it's easy to envision a scenario where digital watermarking becomes a competitive race against hackers. Until a new standard is developed, we can only hope for the best when it comes to new tools like Google's SynthID, an identification tool for generative art, which will continue to get workshopped by developers until it hits the mainstream.
Further reading: Researchers Tested AI Watermarks -- and Broke All of Them
This discussion has been archived. No new comments can be posted.

Researchers Say Current AI Watermarks Are Trivial To Remove

Comments Filter:
  • Guess we've reached an era where cleaning up watermarks is easier than my morning coffee spill. Let's toast to the University of Maryland for stepping up the game, while we're still playing catch-up with hackers. Cheers to the next level of digital hide and seek.
  • The only way to get around this would be to save the prompts into a database and use them after the fact to see what they make, and verify provenance. You won't be able to tell FROM a digital artifact itself where it came from unless they have some kind of crypto signature registered externally. This is like the CIA framing up a laptop to look Russian all over again.
  • The researchers of this paper are apparently all AI and little DSP.

    https://github.com/arpitbansal297/Certified_Watermarks/tree/master

    They did a miserable job of providing code. They should have provided automated tests in their code. Instead, they force people reviewing their work to spend their time guessing.

    My first attack which worked quite well was to perform 5 levels of DWT, then delete blocks in a checkerboard pattern from all but the lowest frequency, run a simple in-paint common to video error conce
  • I can think of a technology that could be tuned to easily remove watermarks in AI... it's AI. What the fuck, idiots
  • Say, by using AI?
  • by Anonymous Coward

    forgive me if I don't play a violin, the less people are tracked the better
     
    ...is a rather absolute stance that should probably acknowledge some nuance, yet the world (our Betters, rather) have insisted on pushing a status quo that resulted in such a stance being mentally drilled in so firmly

    forgive me if everything is such a shitshow that jumping straight to "the less people are tracked the better" is all I can see anymore

  • Artist here. While I can't say every artist understands watermarks, I would say the overwhelming majority of artists understand that removing watermarks is a fairly trivial process with modern software. Artists usually use one of two methods to suppress copyright infringement:

    • The image is posted at full resolution with a watermark. It serves as notice to honest people that the image should not be reposted, and provides proof in cases of copyright infringement that the copying was intentional - which e

"The whole problem with the world is that fools and fanatics are always so certain of themselves, but wiser people so full of doubts." -- Bertrand Russell

Working...