Google has opened its SynthID Detector to everyone in 2026, taking a tool once limited to select partners and putting it within reach of any user who wants to check whether a piece of media carries Google's invisible watermark.

That is the headline fact, and it's a significant one. SynthID is Google's system for embedding imperceptible watermarks (statistical patterns hidden in a file's pixels, audio waveform, or text) into content generated by its AI models. The Detector is the flip side: it reads a file and reports whether that hidden signature is present.

What actually shipped?

The announcement, reported by ALM Corp, confirms the public availability of the Detector itself. It does not carry a launch date more precise than 2026, a price point, or a list of supported file formats. So the honest summary is short: the tool exists, it's open, and the details are thin.

What we know about the SynthID family frames why this matters. Google has positioned watermarking as its main answer to the provenance problem — the question of whether a photo, song, or paragraph came from a machine rather than a person. A detector that only insiders could run was always going to have limited effect. A detector anyone can run changes the arithmetic for newsrooms, platforms, and moderators.

In plain terms: the scanning half of Google's watermarking system is now a public utility, not a private lab.

Who uses this, and what can't it do?

The natural users are the people who already handle unverified media daily: editors checking submissions, platform trust-and-safety teams, and lawyers assembling evidence. For them, a quick yes-or-no on Google's watermark saves a step that previously required a relationship with the company.

But the tool's reach has hard limits, and operators should keep them in mind.

• It only detects watermarks from models that embed SynthID in the first place.
• Content from other AI vendors, or from hand-edited files, will simply read as unwatermarked.
• An absent watermark is not proof a file is human-made — it's an absence of evidence, nothing more.

That last point is where the skepticism belongs. Watermark detectors catch cooperative content. They don't catch the file that passed through a converter, a screenshot, or a competing generator. Any workflow that treats "no watermark found" as "authentic" is building on sand.

The takeaway: useful screening tool, weak standard of proof.

What we still don't know

The public announcement leaves the commercial and technical fine print unset. There's no stated word on rate limits, API access, pricing tiers, or whether the open version matches the capability Google uses internally. Until Google publishes those terms, the practical question for any newsroom or platform is simple: does the free version's accuracy and throughput hold up at production volume?

We'll follow the documentation as it lands. For now, the fact on the record stands — the Detector is open to all comers, and the burden of explaining what it can and can't prove now sits with the people who use it.