Google's SynthID AI Watermarking Adopted by OpenAI, Nvidia, and ElevenLabs

Google has announced that OpenAI, Nvidia, ElevenLabs, and Kakao will adopt its SynthID AI watermarking technology, marking a rare moment of cross-industry cooperation on AI content authenticity. The partnerships mean that AI-generated images, video, and audio from some of the largest platforms will now carry invisible watermarking that can be detected programmatically.

What is SynthID?

SynthID is Google's AI watermarking system that embeds imperceptible digital signatures into AI-generated content. Unlike visible watermarks, SynthID markers are invisible to the human eye and designed to survive common transformations including cropping, compression, and colour adjustment.

According to Google Research director Pushmeet Kohli, SynthID has undergone extensive robustness testing against adversarial attacks. "A technology like this will always be attacked. There was a lot of research that we did in making SynthID robust to different kinds of transformations," Kohli said.

Who is adopting it

The adoption list includes some of the biggest names in AI:

• **Nvidia** will integrate SynthID into its Cosmos world foundation models
• **OpenAI** will use SynthID for GPT-2 image generation output
• **ElevenLabs** will watermark AI-generated audio content
• **Kakao** will apply SynthID across its AI services

Until now, SynthID could only be applied to content generated by Google's own AI models. These partnerships dramatically expand the reach of the technology, potentially creating a de facto standard for AI content authentication.

How you'll check for SynthID

Google is also making SynthID detection more accessible. The watermark status will be checkable through:

• **Circle to Search** on Android
• **Google Lens**
• **AI Mode** in Google Search
• **Gemini in Chrome** — share a tab and ask "Is this AI?"

Google is also expanding its use of the C2PA content provenance standard. Photos from the Pixel 8, 9, and 10 will include C2PA metadata, and the same capability is coming to video recordings. Gemini itself will soon be able to explain a file's provenance based on content labelling, with Chrome and Search getting the same capability in the coming months.

What this means for self-hosted AI

For the self-hosted community, the SynthID standard is a double-edged sword. Open-source image and video generation models—Stable Diffusion, FLUX, and others—do not include SynthID watermarking, meaning content generated on local hardware remains unmarked. This preserves privacy and autonomy but also means AI-generated content from self-hosted systems is indistinguishable from human-created work.

If universal AI watermarking becomes a regulatory requirement, self-hosted operators may eventually need to implement their own watermarking solutions or risk having their content treated as "unauthenticated." For now, however, the open-source ecosystem remains free of mandatory labelling, and the tools covered in our guide to Running Models Locally with Ollama produce fully unmarked output.

Source

**Ars Technica:** https://arstechnica.com/google/2026/05/googles-synthid-ai-watermarking-tech-is-being-adopted-by-openai-nvidia-and-more/