AI-rewritten: This is a summary of an article from Ars Technica, rewritten by AI (Qwen, running locally) to make it easier to read. The facts come from the original article – read it for the full story.

---

Google has launched a new global tool to detect hidden watermarks in AI-generated content. Every image and video from Google’s Gemini models carries an invisible SynthID label, which makes it hard for the content to be mistaken as authentic. While other major AI companies like OpenAI and Nvidia also use similar invisible signals, checking for them was previously difficult. Now, Google says its detector supports watermarks from all its partners, including OpenAI, Nvidia, Kakao, and Apple. Users can visit a new dedicated website to check files for these signatures instead of asking Gemini directly.

To use the tool, you must log in with a supported account, such as those from Google, OpenAI, or Apple. Each user receives a daily usage quota of approximately ten checks for images, videos, and audio. Google engineers told Ars that this limit is necessary to prevent people from using the checks to develop tools that remove watermarks. The system may also lock users out sooner if they check multiple very similar images, which could signal an attempt to bypass the detection workflow.

The new website provides a simple yes-or-no message but does not show specific regions where SynthID pixels appear as internal tools did. Google acknowledges that this system is not perfect because some model providers use different watermarking technologies or none at all. Additionally, open-weight models allow users to run AI on their own hardware without metadata labeling. Despite these limitations, Google states that SynthID remains the best current option for identifying AI fakes in media proliferating across the Internet.