Google's SynthID watermark adopted by OpenAI, Nvidia, and ElevenLabs

Google's SynthID, first demonstrated three years ago, has already been used to label 100 billion images and videos plus 60,000 years of audio. Unlike fragile metadata, SynthID is a digital watermark deeply embedded in pixels or waveforms, making it resistant to compression, cropping, and rotation. Google DeepMind scientist Pushmeet Kohli notes the team worked to make it robust against attacks, and Google claims no public bypass has worked.

Now SynthID is going multi-platform: OpenAI will use it in GPT-2 images, Nvidia in Cosmos world models, and Kakao and ElevenLabs in their AI content. Google is also integrating SynthID detection into Circle to Search, Lens, AI Mode, and Chrome, so users can easily ask “Is this AI?” without opening Gemini. While no public API exists yet to prevent circumvention, Google plans to launch an enterprise detection API under its Gemini Enterprise Agent Platform, allowing trusted partners to flag AI content at scale.