AI · Oct 21, 2026
OpenAI on Google Cloud lets companies spend Google commitments on GPT-6, and that changes procurementGoogle’s global SynthID site now detects watermarks from OpenAI, Nvidia and Kakao, with Apple coming
· Updated · Signal Desk newsroom
Analysis of reporting by Ars Technica. How we work

Google opened a dedicated SynthID website where anyone can check images, video or audio for invisible watermarks, now including partners’ marks. Logins and a daily quota of about ten checks apply, and a negative result does not prove content is real, Ars Technica notes.
Key takeaways
- The site detects SynthID watermarks from Google and partners OpenAI, Nvidia and Kakao; Google says Apple soon.
- Previously, people outside trusted testers could only ask Gemini, which checked only Google’s mark.
- Gemini has stamped SynthID on more than 180 billion images and videos and 240,000 years of audio, Google says.
- Users log in with Google, OpenAI or Apple accounts and get roughly ten checks a day.
- Meta uses different watermarking, and many generators use none.
What changed
Ars Technica reports that Google has opened a dedicated SynthID website where anyone, anywhere, can check an image, video or audio file for the invisible watermark SynthID embeds. The bigger change is coverage: the detector now recognizes the distinct watermarks of Google’s SynthID partners, including OpenAI, Nvidia, Kakao and, Google says, Apple soon. Previously, people outside Google’s trusted tester program could only ask Gemini, and that check looked only for Google’s own mark. TechCrunch’s account details supported file formats (public SynthID site).
SynthID hides its signal in the pixels of images and video and in the waveform of audio. Google says Gemini alone has stamped SynthID onto more than 180 billion images and videos and the equivalent of 240,000 years of audio. The earlier internal tool could highlight which regions carried a watermark; the public site reports only whether SynthID was used to create or edit a file. Google has extended watermarking to other domains, including AI-designed proteins (protein watermarks).
Access limits
Users must log in with a Google, OpenAI or Apple account, and Google describes a daily quota of roughly ten checks. Google engineers told Ars the cap exists to stop people using the detector to build watermark-removal tools, and the system may cut users off sooner if they submit near-identical files. The reasoning holds, but it means journalists and fact-checkers working a breaking story cannot rely on the tool at volume. A single destination covering several large AI providers, delivering a plain answer instead of a conversational one, is still a substantial improvement over scattered company-specific detectors.
What a negative result means
Meta uses different watermarking technology, some providers use none, and open-weight models on personal machines can produce realistic media with no label. Much synthetic content will pass this check unflagged, and Google is posting a disclaimer saying so. The risk is that a screenshot of a no-watermark result circulates as evidence a fake is authentic, turning an anti-deception tool into a way to launder fakes.
Ars points to cryptographic provenance such as C2PA, which labels real media at capture, and notes Google’s Pixel phones already do this. A durable approach likely combines watermarks indicating AI generation with credentials indicating camera origin. This launch improves the first half without settling the second. More in our Google coverage, OpenAI coverage and generative AI coverage.
Who feels it
- Fact-checkers and journalists
- A faster first check, limited by quotas and blind spots.
- Everyday users
- Free verification, if a negative is read as “not found,” not “real.”
- Providers outside SynthID
- Pressure to interoperate or explain why not.
- Platforms
- Cross-provider labeling becomes more practical if access scales.
What to watch
- When Apple watermark support goes live.
- Meta or others joining or interoperating with SynthID.
- Tools that strip SynthID marks.
- Higher-volume access for newsrooms and researchers.
FAQ
Which watermarks does the site detect?
SynthID marks from Google and partners OpenAI, Nvidia and Kakao, with Apple coming, Google says.
Are there usage limits?
Users log in and get roughly ten checks per day.
Does a negative result mean content is real?
No; many generators use other watermarks or none.
Why the quota?
Google says it prevents use of the detector to build watermark-removal tools.
More on OpenAI
AI · Oct 14, 2026
White Circle says newer OpenAI and Google image models are easier to jailbreak than their predecessorsAI · Oct 19, 2026
Microsoft ships Guardian Agents, saying frontier models ‘can’t be trusted’AI · Oct 13, 2026
After months off the frontier, Google ships Gemini 4 Argon with benchmark claims and aggressive pricingAI · Sep 22, 2026
Opus 5.5 hands risky cyber prompts to an older Claude model, and that routing is the real launch story