OpenAI announced new measures on Tuesday intended to make it easier to identify images produced by its artificial intelligence systems. The company said it will integrate Google DeepMind's SynthID invisible watermarking into images created via ChatGPT, Codex, and the OpenAI API, and it is previewing a public verification tool that examines provenance signals.
OpenAI described the decision as part of a broader move to surface information about how media was produced. The firm said it has become a Coalition for Content Provenance and Authenticity Conforming Generator Product, which enables platforms to read and preserve metadata and cryptographic signatures bundled with content produced by OpenAI systems. The C2PA framework is built around metadata and cryptographic signatures that travel with the media.
The SynthID layer from Google DeepMind is an invisible watermarking technology designed to complement metadata-based approaches such as C2PA. OpenAI said the watermarking layer is engineered to survive common transformations, including screenshots and file format conversions, so that provenance signals remain detectable after editing or repackaging.
Alongside the watermarking rollout, OpenAI will provide a public verification utility that inspects uploaded images for provenance indicators. The preview of this tool will identify whether an image was generated using ChatGPT, Codex, or the OpenAI API by checking for both Content Credentials metadata and SynthID watermarks. At launch, the verification utility will only evaluate content produced by OpenAI models.
OpenAI noted that it has applied Content Credentials to images generated by DALL·E 3 since 2024 and later extended that practice to ImageGen and Sora. The company also said it joined the C2PA Steering Committee, the industry group that oversees the open technical standard for content provenance.
Looking ahead, OpenAI indicated plans to participate in cross-industry verification efforts and to support a broader set of content types over time. The company described these as future enhancements, and the initial verification offering remains limited in scope to OpenAI-generated material.
Summary
- OpenAI will integrate Google DeepMind's SynthID invisible watermarking into images generated by ChatGPT, Codex, and the OpenAI API.
- A public verification tool preview will check images for SynthID watermarks and C2PA Content Credentials to confirm OpenAI provenance.
- The verification tool is initially restricted to content generated by OpenAI; broader cross-industry and content-type support is planned in the future.