SynthID Detector открывают всем — он ищет конкретную маркировкуSynthID Detector opens to everyone, with a specific watermark to find

Google announced global access to SynthID Detector in English on October 7. The tool checks images, video and audio made with AI from Google and partners. The announcement names OpenAI, NVIDIA and Kakao, with Apple listed as a future addition. An earlier version targeted media professionals. The development is broader access to verification, rather than an infallible method for identifying every artificial image. Keeping future tense for a partner that has yet to join the announced support is part of describing the release accurately.

Google DeepMind's technology description explains the mechanism: SynthID embeds an imperceptible watermark when content is generated. For images and video, the company describes resilience to modifications including cropping and compression. That is a watermark property under described conditions, rather than a guarantee after every conceivable transformation. The technology also supports text, but today's public detector announcement lists three media types. The wider SynthID capability set should not be confused with this particular verification interface. Readers need to know which function is becoming available now.

Our reading is that a detector addresses a narrower question than the familiar “is this real?” A found watermark can provide information about a supported generator's involvement. Whether the depicted event is true remains a different question. A synthetic illustration can honestly explain an idea; an ordinary photograph can carry a false date or caption. These examples illustrate the distinction between file origin and publication meaning, rather than describe incidents discovered by this publication. Failure to find a signal cannot by itself establish human authorship: the check concerns a particular watermarking technology, rather than every possible way to produce media.

For editorial use, we would preserve the check result alongside the original file, time and a link explaining the tool. We would then examine the caption, location and publication circumstances separately. That sequence keeps the conclusion within what has actually been established: one particular file was checked by a particular service. A retelling without those details can turn a technical clue into an excessively broad claim. We uploaded no user materials and performed no comparative detector evaluation. This story concerns the new announcement and how to interpret its stated function without expanding its scope.

Why it matters

Broader access brings marked-media verification to a wider audience. Its value also depends on how the result is communicated. Our conclusion is to keep a detected origin signal distinct from established truth about an event. The new tool can then help examine publication context while leaving the other questions visible, rather than collapsing them into a single answer.