Story
October 7, 2026
Google Opens Its AI Detector to All—But It Can’t See Every Fake
Google is pitching its public SynthID detector as a simpler way to verify whether media came from major AI systems, while the tool’s limits underscore a harder truth: watermark checks can only catch content whose creators chose to leave a trace.
Google first offered its SynthID Detector to media professionals last year, giving them a way to scan images, video and audio for the company’s near-invisible AI watermark. The system is now open to anyone who signs in at SynthID.com.1
The wider rollout marks a significant expansion. Google says the detector can now recognize watermark signals from participating partners, including OpenAI, Nvidia and Kakao, with Apple due to join. Demis Hassabis amplified the announcement, describing the service as a way to check content made with Google AI or industry-partner tools.
2
For users, the shift removes a cumbersome earlier route: outsiders had previously needed to ask Gemini to assess a file. The new site instead returns a straightforward indication of whether supported SynthID technology was used to create or edit the media. Google says Gemini’s output alone has carried SynthID across more than 180 billion images and videos, as well as 240,000 years of audio.3
But the apparent simplicity masks a sharp boundary. The detector is not a universal AI-fake test; it can only find the marks it has been built to recognize. Other providers use different watermarking systems, while some use none at all. Open-weight models run locally can also produce realistic material without even basic metadata labels.3
Access is limited, too. Users must log in and receive roughly 10 checks a day, a restriction Google says is intended to stop bad actors from using the service to develop watermark-removal techniques.3
That leaves SynthID as a useful checkpoint rather than a final verdict: a positive result can identify some AI-made media, but a clean scan cannot prove a file is real.