Google’s AI Checker Goes Global, but Its Blind Spots Remain

Google’s AI Checker Goes Global, but Its Blind Spots Remain
Google first offered its SynthID Detector to media professionals last year, using near-invisible markers embedded in AI-generated images, video and audio. It has now opened the service worldwide: anyone with an account can upload material to SynthID.com and check whether it bears Google’s watermark.

The expansion is more than a consumer release. Google says the detector can now recognize marks from industry partners, turning what had been a fragmented set of checks into one portal. In a post amplified by Google DeepMind chief Demis Hassabis, the company said the tool can check content made with Google AI as well as systems from OpenAI, Nvidia and Kakao, with Apple “coming soon.”

That matters because Google’s earlier detector could identify only its own version of SynthID—even when another company’s content carried a compatible label. The new site gives users a simple finding on whether supported SynthID technology was used to create or edit a file, rather than the more detailed internal view that could highlight marked regions. Google says access is capped at roughly 10 image, video or audio checks a day, a safeguard intended to make it harder for people to develop watermark-removal techniques.

Still, a clean result is not proof that media is real. Providers such as Meta use different systems, while other model makers use no watermarking at all. Open-weight models can also generate convincing content locally without even basic metadata labels. And if a SynthID mark is removed, the checker may have nothing to find.

So the launch is a useful verification tool, not a universal lie detector. The broader alternative, as the debate over provenance intensifies, is to cryptographically label authentic media at the moment it is created—flipping the problem from trying to catch every fake to proving what is genuine.

https://foxvector.com