Google Opens Its SynthID Detector to Everyone: What It Can and Can’t Tell You

Google has opened its SynthID Detector to the public. Since 7 October 2026, anyone can check whether an image, video or audio file carries a SynthID watermark from Google or one of its partners, including OpenAI, NVIDIA and Kakao, with Apple “soon” [1].

That is confirmed in a Google blog post by Pushmeet Kohli of Google DeepMind [1]. It is also more limited than “AI detector” suggests. It looks for one family of invisible watermarks, so it can say “yes, this was made or edited with a supported tool”. It cannot prove that something without a mark was made by a human.

What has changed?

Google says it introduced an early version of the SynthID Detector last year for media professionals. Now it is “expanding access to everyone”, available globally in English from 7 October [1].

The bigger change is coverage. Google says the new detector lets anyone check if an image, video or audio file “was made with AI from Google or our partners, including OpenAI, NVIDIA, Kakao, and soon, Apple” [1].

Ars Technica explains why that matters: Google’s previous detector only checked for Google’s own version of SynthID, so a ChatGPT image carrying SynthID would not have been flagged. The new tool checks for the distinct watermarks of all of Google’s SynthID partners in one place [2]. The Verge reports that the tool is at synthid.com and requires a login [3].

What is SynthID, in plain words?

SynthID is a watermark you cannot see or hear. Google describes it as an “imperceptible” mark across images, video and audio [1]. Ars Technica describes it as a signal encoded in the pixels of AI images and video and in the waveform of AI audio [2].

The idea is that a tool adds the mark at the moment it creates or edits a file. A matching detector can later look for it. If the detector finds it, that is strong evidence the file came from, or passed through, a tool that adds that mark.

Google says it has watermarked more than 180 billion images and videos, and 240,000 years of audio, since launching SynthID in 2023 [1]. It adds that its built-in verification features in Search, the Gemini app and Chrome now regularly handle more than one million requests a day [1].

How does the public tool work?

According to Ars Technica [2]:

• You must log in. Google, OpenAI and Apple accounts are supported.
• Each user gets, in Google’s words, “a daily usage quota of approximately 10 image, video, and audio checks”.
• The result is a simple message saying whether SynthID was used to produce or edit the media. The older internal tool highlighted which regions of an image carried the mark; the public site does not.

Google engineers told Ars the usage limit exists to stop people using repeated checks to build tools that strip SynthID out [2]. That is a reasonable trade-off: a detector anyone can query without limit is also a test bench for removing the mark.

What a “no” means, and what it doesn’t

This is the most important point for readers.

A positive result is meaningful. It says a supported company’s tool made or edited the file.

A negative result is weak. It only means no supported SynthID watermark was found. Ars Technica lists why that is not proof of authenticity [2]:

• Some AI providers use different watermarking technology. Ars names Meta as an example.
• Some use no watermark at all.
• Open-weight models that people run on their own hardware can produce realistic content with no label of any kind.
• Someone may find a way to remove SynthID from existing content.

Ars reports that the website itself will carry a disclaimer saying that plenty of AI content will not be identified [2]. Treat the tool as one check among several, not a verdict.

Why it matters

Two things stand out.

First, a shared standard is forming. Having OpenAI, NVIDIA and Kakao already supported, and Apple promised, means one free check now covers content from several of the biggest image, video and audio generators [1][2]. For journalists, teachers and anyone who sees a suspicious clip, that is a real improvement on asking each company separately.

Second, it fits a wider shift from voluntary labels towards routine provenance. In the past week TSN has covered OpenAI’s plan to watermark ChatGPT text in the EU and DeepMind research on hiding a made-by-AI tag inside designed proteins (see Related below). Note the difference: Google’s new public detector checks images, video and audio [1]. It is not described as a text checker.

Ars also points to the other half of the problem. Instead of marking fakes, technologies such as C2PA can cryptographically label real photos, video and audio at the moment of capture [2]. The two approaches complement each other.

What this does not prove

• That unmarked content is human-made. Many AI tools add no SynthID mark, and open-weight models may add nothing at all [2].
• That the watermark cannot be removed. Google limits checks precisely because removal is a risk [2].
• When Apple support arrives. Google says only “soon” [1].
• That it checks text. The announcement covers image, video and audio files [1].
• Accuracy rates. Google’s post does not publish false-positive or false-negative figures for the public tool.

The Bottom Line

Google’s SynthID Detector is now open to everyone, and it checks for watermarks from Google, OpenAI, NVIDIA and Kakao, with Apple to come [1]. That makes it the most useful single place to ask “was this made with one of the big AI tools?” for images, video and audio. Use a “yes” as strong evidence. Treat a “no” as “not marked by a supported tool”, nothing more [2].

Related on TSN: OpenAI’s textGrain: invisible watermarks for EU ChatGPT text · The Invisible Signature: Google DeepMind Has Started Watermarking AI-Designed Proteins

Sources

1. Pushmeet Kohli, “We’re making it easier to identify AI-generated content globally” (page title “Google expands SynthID Detector for AI content”), Google blog, 7 October 2026. https://blog.google/innovation-and-ai/models-and-research/google-deepmind/synth-id-ai-content/
2. Ryan Whitwam, “Google rolls out improved SynthID AI content detector, now available globally,” Ars Technica, 7 October 2026. https://arstechnica.com/ai/2026/10/google-rolls-out-improved-synthid-ai-content-detector-now-available-globally/
3. Emma Roth, “Google’s AI content detector is rolling out globally,” The Verge, 7 October 2026. https://www.theverge.com/tech/1006640/google-ai-content-detector-launch