An AI watermark can tell you something useful. It cannot settle the whole story.

The update is access

On October 7, Google expanded SynthID Detector to everyone, globally in English. The service checks images, video and audio for SynthID watermarks associated with supported AI systems. Google previously offered an early detector to media professionals.

SynthID’s underlying mechanism is important: participating systems embed an imperceptible watermark when content is generated. The detector looks for that signal. Google’s technical overview describes watermarking and identification; this is not the same as a universal test for every possible AI tool or a fact-check of the claim accompanying a file.

Source: Google: Expanded access to SynthID Detector ↗ · Google DeepMind: How SynthID works ↗ · Penguin Random House: The Coming Wave ↗

Our take: keep two questions separate

One question is how a piece of media was made. Another is whether the story being told about it is accurate. A generated illustration can responsibly explain an idea when it is labeled. A real photograph can still be attached to the wrong event. Knowing the production method helps, but it does not answer when, where or why something happened.

A detected watermark is a useful reason to describe supported AI involvement. An absent signal should not become a declaration that a file is authentic. Our interpretation follows from the detector’s scope: a tool looking for one family of signals cannot establish the origin of every file that lacks those signals.

For a broader discussion of the risks and governance questions raised by powerful technologies, The Coming Wave (affiliate link) is optional context. You do not need the book—or a detector result—to start checking whether the original source supports a post’s claim.

As an Amazon Associate I earn from qualifying purchases. This optional recommendation is not a claim of hands-on testing.

A practical pause before sharing

Save the claim in your own words first. For example, a clip may be presented as showing a newly announced product doing a specific task. Then look for the original publisher, a dated announcement and any explanation of what the demonstration represents. Separate a concept video from evidence of a shipping feature.

If you use a verification tool, write down exactly what it reported. Avoid translating an uncertain or limited result into a stronger conclusion. Keep the original link with your note so another reader can inspect the same context. If the origin remains unclear, say so instead of filling the gap with a guess.

Use files you are entitled to share with the verification service. A request to upload media deserves its own privacy judgment, especially for personal recordings or material involving someone else. Check the service’s current terms and handling of uploads before sending private content.

Make the correction easy

When you post your own AI-created artwork, label it at the point where people encounter it. Do not rely on a subtle filename or a disclosure tucked away on another page. If someone mistakes an illustration for documentary evidence, correct the impression clearly.

For an uncertain post, it can be enough to withhold the share until you find a primary source. That is a small action with a real benefit: you avoid adding confidence to a claim you have not established. We have not tested SynthID Detector or measured its accuracy, and our suggested verification routine is not a guarantee that misinformation will always be caught.