No, You Can't Verify a Video Is AI-Generated Using Google's SynthID — Here's Why That Claim Falls Apart
“The viral video contains AI-generated visual content detectable by Google's SynthID watermark technology”
The argument in brief
A viral claim asserts that a video was confirmed as AI-generated through Google's SynthID watermark technology. This is unverifiable. SynthID only works on content made by Google's own AI tools, and its detection system is not publicly accessible — meaning no outside person can actually run that check.
Why it spread
People are genuinely worried about AI deepfakes, and that fear makes them want clear answers. A claim that invokes Google's name and a technical-sounding system like 'SynthID watermarking' feels authoritative and reassuring — like someone with real tools did the hard work. That borrowed credibility is exactly what makes it spread before anyone stops to ask whether the check was ever actually possible.
The claim is that a viral video has been identified as AI-generated because it carries a detectable Google SynthID watermark. The verdict is simple: this cannot be confirmed or denied by anyone outside Google, and the claim almost certainly misrepresents how SynthID actually works.
SynthID is real. Google DeepMind developed it, and a peer-reviewed paper published in Nature in 2024 confirms it can embed and detect invisible watermarks in AI-generated images, audio, text, and video. So the technology exists — but that's where the accuracy of the claim ends.
Here's the critical problem: SynthID only watermarks content created by Google's own AI systems, like Imagen and Veo. If a video was made with Midjourney, Sora, Runway, or any other non-Google tool, it would carry no SynthID watermark at all. The MIT Technology Review has reported that no single watermarking system covers all AI-generated content, and the Content Authenticity Initiative, an industry coalition including Adobe, confirms SynthID's scope is limited to Google's own outputs.
Even if a video were made with a Google AI tool, the detection system is proprietary and not publicly available. As Snopes and other fact-checkers have noted, ordinary users — and most journalists — simply do not have access to Google's SynthID detection API. No one outside Google can run this check. A claim that someone has done so is either mistaken or fabricated.
This kind of misinformation is worth watching for because it wraps a false conclusion in real, credible-sounding technology. The name 'Google' and the word 'watermark' make it feel scientific and settled. When you see a claim that a specific piece of content has been verified by a proprietary corporate tool, ask one question first: did the person making the claim actually have access to that tool?
Sources
- Google DeepMind - SynthID Overview
SynthID is a real watermarking tool developed by Google DeepMind that embeds imperceptible watermarks into AI-generated images, audio, text, and video. However, detection requires access to Google's proprietary detection tools and is not publicly available for third-party verification.
- Google DeepMind SynthID Research Paper (Nature, 2024)
The peer-reviewed paper confirms SynthID's technical capability to watermark and detect AI-generated content, but notes the system works only for content generated by Google's own AI systems (e.g., Imagen, Veo). Content generated by other AI tools would not carry SynthID watermarks.
- MIT Technology Review - AI Watermarking Limitations
Experts note that AI watermarking technologies, including SynthID, are not universally applied across all AI generation platforms, meaning the absence or presence of a SynthID watermark cannot confirm or deny whether any given video is AI-generated.
- Snopes - AI Detection Claims
Snopes and similar fact-checkers have repeatedly noted that claims about specific videos being verified as AI-generated via proprietary watermark tools are rarely substantiated, as public users do not have access to SynthID's detection API.
- Content Authenticity Initiative (CAI) / Adobe
Industry coalitions working on provenance standards note that no single watermarking system covers all AI-generated content, and SynthID specifically only applies to outputs from Google's own generative AI models.