Google has rolled out an improved version of its SynthID AI content detector, now available globally. The tool detects watermarks from supported companies including Google, OpenAI, and Apple. Users must log in, and each gets a daily quota of approximately 10 image, video, and audio checks.
The login requirement limits usage to prevent people from developing SynthID removal tools. Google engineers told Ars that similar limits existed in previous versions. The exact quota may vary, as the system might lock out users checking multiple similar images, which could signal attempts to bypass the system.
Expanding access is positive, offering a single destination to check media for AI signatures. However, SynthID only covers Google's partners. Other providers like Meta use different watermarking, and some use none. Open-weight models can generate realistic AI content without metadata, meaning much AI content will go undetected. The website includes a disclaimer about this.
Despite efforts, SynthID may still be bypassed. The reality of proactive watermarking is imperfect. An alternative is cryptographic labeling like C2PA, which marks real content as created. Google's Pixel phones are among few devices doing this. For now, SynthID remains the best available tool for identifying AI fakes.
Source: Ars Technica · Summarized by HeadlinesBriefing