Key facts
- Google's SynthID tool, which detects AI-generated content watermarks, is now available globally.
- Users must log in to use the tool, with a daily quota of about 10 checks.
- The daily quota is limited to prevent the development of tools to bypass SynthID.
- SynthID supports watermarks from Google, OpenAI, and Apple.
- Some AI content may not be identified as AI by SynthID due to different watermarking technologies or lack thereof.
Google has made its SynthID tool for detecting AI-generated content watermarks available globally. The tool requires users to log in and imposes a daily quota of approximately 10 checks for images, videos, and audio. This limitation is in place to prevent the development of methods to circumvent SynthID's detection capabilities, according to Google engineers.
While the expanded access is seen as beneficial for providing a centralized destination to identify AI-generated media, the tool is not foolproof. Some AI model providers employ different watermarking techniques, such as Meta, while others use no watermarks at all. Additionally, open-weight models can be run locally, producing content without metadata labeling, which SynthID would not detect. The tool will include a disclaimer acknowledging these limitations.
Google acknowledges that SynthID is not a perfect system, given the rapid proliferation of AI content. The company suggests that cryptographic technologies like C2PA, which label authentic media at creation, might be a more robust long-term solution. Google's Pixel phones are noted as an example of devices implementing such technology. For now, SynthID is presented as the most effective available method for identifying AI fakes.
