Google has opened its SynthID Detector to the public worldwide, with English-language access starting October 7, 2026. The web-based verification tool examines images, video and audio for invisible AI watermarks used by Google and several partner companies. It offers another way to investigate the origins of digital media, but it is not a universal test for whether a file was made with artificial intelligence.
What changed on October 7
In Google’s October 7 announcement, DeepMind executive Pushmeet Kohli said the SynthID Detector is now available globally in English. The detector was introduced in a more limited form for media professionals in 2025. The broader release brings that verification workflow to general users, who can submit supported media for watermark checks. Google says it has applied SynthID to more than 180 billion images and videos and audio material totaling the equivalent of 240,000 years since the technology began in 2023. Those volumes refer to media watermarked through participating systems; they should not be mistaken for the number of fakes identified by investigators or the number of files uploaded to the new detector.
How the SynthID Detector works
SynthID is a digital watermarking system designed to place an imperceptible signal into AI-produced media. For an image, the marker can be embedded within pixel patterns; video and audio require corresponding approaches for their respective data. As Google DeepMind explains, the goal is to make a marker detectable through a compatible verification process without visibly altering the content. A user supplies a file to the SynthID Detector, and the service checks for supported watermark signals. This is different from a classifier that guesses whether a photo looks artificial or whether a voice sounds synthetic. A watermark can convey evidence about a participating generation process, while appearance-based judgments rely on statistical patterns and can be unreliable.
Which AI services and files it supports
Google says the expanded detector can examine images, video and audio from systems that use compatible watermarks. Its October announcement names Google, OpenAI, NVIDIA and Kakao as participating companies, with Apple support described as coming soon rather than already active. The initial public interface is in English. Google’s documentation also describes related verification experiences in Search, Gemini and Chrome; the newly public portal is a dedicated route for users who prefer to check a particular file. Coverage varies depending on whether a specific producing model embeds a recognized marker, and broad brand support should not be interpreted as proof that every output from every product of those companies is watermarked.
What a watermark result can and cannot prove
A positive watermark result provides useful evidence that media was produced or processed with a compatible AI service. It does not, by itself, verify who uploaded the file, when the depicted events happened, or whether surrounding captions are truthful. Conversely, failure to detect a watermark does not establish that a photo or recording is authentic. The material might have been made by an AI system that does not support SynthID, may use an unsupported format or may have undergone transformations affecting recoverability. The detector also does not automatically distinguish a wholly generated scene from legitimate footage that has received some AI editing. Editorial teams should treat results as one piece of a provenance assessment, not a definitive true-versus-false verdict.
What this means for media verification
Public access reduces the friction that once forced many nonprofessionals to rely on limited-access tools or verification features embedded within other products. Journalists, educators and researchers can use watermark checks alongside source interviews, original-file inspection, reverse searching and corroboration from independent evidence. For platforms, the announcement illustrates a move toward interoperable content-provenance signals spanning multiple model developers. But the utility of such a system depends on how consistently watermarks are embedded, preserved and recognized across products. Google reports more than one million daily requests through its existing verification features, suggesting substantial interest in provenance checks. The October 7 release extends access to a dedicated portal while leaving the fundamental limits of watermark-based detection intact.
