Google's AI Detector Is Open to Everyone, but It Finds Watermarks, Not Fakes

My verdict: useful, free, and much narrower than the headline suggests. A clean result does not prove that something is human-made.

Google has opened its SynthID Detector website to the public worldwide. You upload an image, video or audio file and it checks for the invisible SynthID watermark, or for C2PA credentials. It is free, but you have to sign in with a Google, OpenAI or Apple account. Per Engadget, it covers content from Google, OpenAI, NVIDIA and Kakao, and support for Apple Intelligence images is listed as coming soon.

The limits matter more than the launch. This is not a model that looks at a picture and guesses whether an AI made it. It looks for a mark that participating companies add when their tools generate content. Content from a tool that does not add the mark will not be flagged, and Google itself says watermarks can be removed and that the tool cannot tell fully generated content from content that was only edited with AI.

I work with text and image generation every day, so I can mostly tell the difference by eye. Even so, I sometimes see a video and cannot work out whether it is real, only that something feels wrong. Human judgment makes mistakes, and expecting any app to get this right on the first attempt would be too much. Still, a checker built only for this job could turn out to be more reliable than a human guess. That is my opinion, and it applies only to content that carries a watermark. For anything else the tool has nothing to find.

How I would use it: as one signal. A positive result is informative. A negative result tells you very little. Try it on something you already know the origin of, and read what Google says it cannot do before you trust the answer. The sources are linked below.