Google's SynthID watermark is hard to break, but it doesn't solve AI disinformation
Ars Technica AI · Ryan Whitwam · 2026-07-29
Google's SynthID watermarking technology is technically robust against manipulation but may be insufficient as a disinformation defense given that generative AI has already produced over 100 billion images, matching 149 years of human photography in under two years.
Extraction
Topics: ai-watermarkingcontent-authenticationdisinformationgenerative-aigoogle-synthid
Claims
- Generative AI produced 1.5 billion images in 18 months, matching the total volume of images humanity created in 149 years of photography.
- Google's generative tools have been used to create more than 100 billion AI images and videos in just a couple of years.
- Google's SynthID watermark is technically hard to remove or circumvent, giving it strong provenance-tracking properties.
- Despite SynthID's technical robustness, labeling AI content may be a losing strategy given the unprecedented scale and diversity of AI media generation.
Key quotes
Starling Lab...estimates that it took until 1975—149 years after the invention of the camera—for humanity to create 1.5 billion images. It took generative AI just 18 months to do the same.
Google announced at I/O that its tools had been used to create more than 100 billion AI images and videos in just a couple of years, and Google is far from the only source of AI content.