OpenAI adopts Google's SynthID watermarking
- OpenAI said on May 19 it added Google’s SynthID watermarking to images from ChatGPT, Codex and its API, alongside C2PA provenance metadata. (openai.com) - Google said on May 19 that OpenAI, Kakao and ElevenLabs were adopting SynthID, after Google had already watermarked 100 billion images and videos. (blog.google) - OpenAI’s public verifier is live now at its research preview, where users can upload images and check for C2PA or SynthID signals. (openai.com)
OpenAI said on May 19 that it is adding Google’s SynthID watermarking to images generated through ChatGPT, Codex and its API, widening a content-authenticity effort that also includes C2PA metadata and a new public verification tool. (openai.com) Google, in posts tied to its I/O announcements the same day, said OpenAI, Kakao and ElevenLabs were adopting SynthID for more AI-generated content. OpenAI described the move as part of a “multi-layered” provenance system, while Google framed it as broader industry adoption of interoperable transparency tools. (blog.google) The announcements landed as AI companies and platforms face growing pressure to show how generated media can be identified after it spreads across the web. (openai.com) ### Which OpenAI outputs are getting Google’s watermark? OpenAI’s May 19 post said images generated with ChatGPT, Codex and the OpenAI API now include both C2PA Content Credentials and SynthID watermarks. The company said C2PA carries provenance metadata about how media was created or edited, while SynthID embeds an invisible signal directly into the image itself. OpenAI’s help documentation, updated the same day, said the two systems are meant to reinforce each other: metadata can carry richer context, and watermarking can survive some transformations when metadata does not. The company said the watermark may persist through edits such as screenshots, while metadata can be stripped. (openai.com) ### What exactly did Google announce about OpenAI, Kakao and ElevenLabs? Google said on May 19 that “companies like OpenAI, Kakao and ElevenLabs are bringing SynthID technology to more of their AI-generated content.” In Sundar Pichai’s I/O keynote transcript, Google also said Nvidia had signed on to SynthID last year and that OpenAI, Kakao and ElevenLabs were adopting it “today.” (openai.com) Google’s blog post said SynthID has been integrated across its own generative media systems for three years and has already watermarked more than 100 billion images and videos, plus 60,000 years of audio. (help.openai.com) The company said that scale is part of a wider push to make verification available across Search, Gemini, Chrome, Pixel and Cloud. ### Why pair SynthID with C2PA instead of using one system? OpenAI said C2PA conformance makes its provenance signals easier for other tools and platforms to recognize, preserve and pass along. The company said it recently became a C2PA Conforming Generator Product, a technical step aimed at making attached provenance data more portable across services. (blog.google) Google and OpenAI both described SynthID as a separate layer. OpenAI said the watermark is embedded in the image itself and is designed to stand up to cropping, filters and lossy compression, while Google said robust, interoperable tools are needed because media moves across many platforms. (blog.google) ### How does the new verification tool work in practice? OpenAI’s research preview lets users upload a PNG, JPG or WEBP image and check for provenance signals associated with OpenAI tools. The company said the verifier looks for supported C2PA metadata and SynthID watermarks and is designed for images generated with ChatGPT, the OpenAI API or Codex. (openai.com) The tool does not judge whether an image is accurate or misleading. OpenAI said a detected signal indicates the image likely originated from its tools, but does not confirm how the image was later used, edited or presented. (openai.com) ### What else did Google say is coming next? Google said verification for C2PA Content Credentials is rolling out in the Gemini app starting now and will come to Search and Chrome in the coming months. The company also said SynthID verification in the Gemini app has already been used 50 million times globally, and that Search support starts now while Chrome support is due in the coming weeks. (openai.com) OpenAI’s next visible step is the public verifier it launched in preview on May 19. Google’s next platform rollouts, according to its blog, are Chrome in the coming weeks for SynthID checks and Search and Chrome in the coming months for C2PA verification. (openai.com) (blog.google)