Google SynthID Detector Expands: What an AI Watermark Can Tell You

SynthID Detector: the confirmed expansion

Google announced on October 7 that SynthID Detector is available globally in English. The tool checks images, video and audio for supported AI watermarks. Google names its own services and partners including OpenAI, NVIDIA and Kakao; Apple is described as a forthcoming addition.

Google’s announcement is the primary source. Check the tool’s current support and instructions before uploading a file. We have not independently tested its detection performance.

This expansion makes a provenance signal easier to inspect. It does not make a watermark check a complete fact-check of everything shown or said in a file.

AI detection and factual verification are different questions

Our analysis: a supported watermark can help establish that media passed through an AI generation system. It does not establish that the caption is accurate, that the depicted event happened or that the person sharing it is trustworthy.

Likewise, the absence of a supported watermark does not prove that a file is an authentic camera recording. The announcement describes supported systems, rather than universal coverage of all editing and generation tools. Treat the result as one piece of evidence.

An illustrative scam might use an ordinary photograph with a fabricated caption claiming a company has endorsed an investment. The photograph’s origin would not settle the endorsement claim. Verify the claimed announcement through the company’s independently located website.

How to check suspicious media without overclaiming

1. Find the earliest identifiable source and check the publication date.
2. Preserve the original file and context when appropriate.
3. Use a supported detector according to its instructions.
4. Record exactly what the result says, including uncertainty.
5. Verify the consequential claim with primary evidence.

Avoid uploading private recordings or confidential documents without reviewing the service’s data handling. If a message pressures you to pay, sign in or install software, pause that action while checking its origin.

For suspicious requests, use our independent verification guide. Our source-checking analysis explains why professional presentation and repeated coverage are weak substitutes for traceable evidence.

Interpret the result without overstating it

This table is our interpretation framework, not a report of detector testing. Record the actual result and tool instructions rather than converting uncertainty into a yes/no authenticity label. The October 7 announcement describes supported Google and partner media; it does not claim coverage of every generation or editing system.

Sources checked October 9, 2026. Update: Added a quick answer and a result-interpretation table distinguishing watermark provenance from factual verification.