Super important for Google as a search engine so they can filter out and downrank AI generated results. However I expect there are many models out there which don’t do this, that everyone could use instead. So in the end a “feature” like this makes me less likely to use their model because I don’t know how Google will end up treating my blog post if I decide to include an AI generated or AI edited image.
The EU didn't define any specific method of watermarking nor does it need to be tamper resistant. Even if they had specified it though, it's easy to remove watermarks like SynthID.
I have been curious about this myself. I tried a few basic stenography detection type tools to look for watermarks but didn’t find anything. Are you aware of any tools that do what you are suggesting?
> Today, we are putting a powerful verification tool directly in consumers’ hands: you can now upload an image into the Gemini app and simply ask if it was generated by Google AI, thanks to SynthID technology. We are starting with images, but will expand to audio and video soon.
Re-rolling a few times got it to mention trying SynthID, but as a false negative, assuming it actually did the check and isn't just bullshitting.
> No Digital Watermark Detected: I was unable to detect any digital watermarks (such as Google's SynthID) that would definitively label it as being generated by a specific AI tool.
This would be a lot simpler if they just exposed the detector directly, but apparently the future is coaxing an LLM into doing a tool call and then second guessing whether it actually ran the tool.
By anybody's AI using SynthID watermarking, not just Google's AI using SynthID watermarking (it looks like partnership is not open to just anyone though, you have to apply).
