Researchers Find Flaws in AI Watermarking Tools Meant to Detect Fake Images

Current AI watermarking methods are easily defeated by researchers like Feizi. He states "We don't have any reliable watermarking at this point."
Feizi's study shows how attackers can both remove AI watermarks and add fake watermarks to human-generated images.
Other researchers agree watermarking has flaws, but some still think it has a role combined with other detection methods.
Companies like Google are hedging bets on watermarking tools like SynthID, admitting they are "not foolproof."
Feizi is skeptical watermarking is a good focus for companies, stating "Perhaps we should get used to the fact that we are not going to be able to reliably flag AI-generated images."

arstechnica.com

Relevant topic timeline:

8/29/2023

AI images are getting harder to spot. Google thinks it has a solution.

Google has announced a new tool, called SynthID, which embeds a digital "watermark" into AI-generated images, making it harder to spread fake images and disinformation.

9/12/2023

Ricketts proposes bill to watermark AI content for transparency

US Senator Pete Ricketts is introducing a bill that would require watermarks on AI-generated content in order to provide transparency to consumers and differentiate between real and AI-generated information.

10/2/2023

Researchers Reveal Limitations of AI Image Watermarking for Detecting Fakes

The reliability of digital watermarking techniques used by tech giants like Google and OpenAI to identify and distinguish AI-generated content from human-made content has been questioned by researchers at the University of Maryland. Their findings suggest that watermarking may not be an effective defense against deepfakes and misinformation.

10/4/2023

Microsoft Integrates OpenAI's DALL-E 3 into Bing - Users Get Free Access But Experts Warn of AI Risks

Microsoft has integrated OpenAI's DALL-E 3 model into its Bing Image Creator and Chat services, adding an invisible watermark to AI-generated images, as experts warn of the risks of generative AI tools being used for disinformation; however, some researchers question the effectiveness of watermarking in combating deepfakes and misinformation.

10/15/2023

AI Falsely Labels Real War Images as Fake, Raising Concerns Over Reliability of Detection Tools

Some AI programs are incorrectly labeling real photographs from the war in Israel and Palestine as fake, highlighting the limitations and inaccuracies of current AI image detection tools.

Topics

Posted 10/3/2023, 3:01:55 PM