9 October 2026: Google has opened its SynthID Detector website to the public worldwide in English, giving more people a way to check for a particular kind of hidden watermark in AI-generated images, video and audio. The move was announced on October 7, 2026. It makes a previously limited verification tool more accessible, but it does not establish whether every suspicious file was made by artificial intelligence.
Key points
- Google made its SynthID Detector publicly available worldwide in English on October 7, 2026.
- The website can check supported image, video and audio watermarks associated with Google, OpenAI, Nvidia and Kakao systems.
- A negative result is not proof that media is authentic, because the detector cannot recognise every AI-generated file.
What the detector can check
Google says SynthID embeds a signal into the output of participating AI systems in ways intended to be difficult for viewers to notice. The new SynthID Detector lets users inspect supported media for that signal. The company’s October 7 announcement names existing partners OpenAI, Nvidia and Kakao, and says Apple integration is planned for a later date. The statement should not be interpreted as proof that every image from every product offered by those companies carries a compatible watermark.
Google says its watermarking technology has been applied to more than 180 billion images and videos and audio totalling 240,000 years in duration. Those are company-reported historical usage measures, not results independently checked in this article. The public tool covers image, video and audio files; the announcement does not claim it can examine all AI-generated text or determine whether a news story is factually correct.
Practical limits
Independent reporting by Ars Technica says the website requires sign-in and that checks are subject to an approximate daily usage quota. Readers should consult the current site for limits rather than assume unlimited access. The published reports do not constitute this newsroom’s hands-on testing of the service.
The major limitation is coverage. Some AI models do not use SynthID at all, and files can be recompressed or altered in ways that affect evidence. A result showing no compatible watermark therefore cannot authenticate a photo, prove an audio recording is genuine or establish a video’s origin. Conversely, the presence of a supported watermark can indicate use of participating AI tools to create or edit content, but it does not by itself tell viewers whether every part of a file is synthetic.
Why it matters
The broader availability gives journalists, researchers and ordinary internet users an additional clue when checking material circulating online. It should be one component in a verification process that also examines the original uploader, publication date, location, context and independent evidence. A detector is not a substitute for reporting or for asking whether a claim in a video is true.
The October 7 change differs from our October 8 coverage of Microsoft’s new hardware. Find related reporting in the Technology section.
Sources
- Google DeepMind announcement, 7 October 2026
- Ars Technica, 7 October 2026
- The Next Web, 7 October 2026
Representative image; not a screenshot of the SynthID Detector or a photograph of Google’s release. Laptop-use stock photograph by Mikhail Nilov, Pexels, under the Pexels License. Image was reused because no sufficiently relevant story-specific image with verified reuse rights was found and a separate new stock upload was safety-rejected.
Editorial disclosure: Original AI-assisted drafting checked against Google’s dated primary announcement and contemporary independent reporting. The author did not log into or experimentally evaluate the service.
Leave a Reply