Advancing content provenance for a safer, more transparent AI ecosystem

OpenAI has strengthened its multi-layered provenance verification system that combines Content Credentials with SynthID. Having recently become a generator product conformant with the C2PA standard, this allows the provenance signals that have been attached to DALL·E 3, ImageGen, and Sora since 2024 to be better read, preserved, and passed along by other tools and platforms.

C2PA metadata carries information about how content was created and edited, but it can be lost during transformations such as uploading, downloading, resizing, or taking screenshots. To address this, OpenAI has partnered with Google to embed an invisible SynthID watermark into images, starting with images generated in ChatGPT, Codex, and the OpenAI API.

OpenAI has also released a public verification tool in preview. This is a follow-up to the image detection classifier research preview released in 2024. Users can check whether an uploaded image was generated with OpenAI tools, and the tool checks both Content Credentials and SynthID together, showing metadata if a signal is present. If no signal is found, it does not make a definitive claim about whether the image was generated, and for now it is limited to images generated by OpenAI. Going forward, OpenAI plans to expand cross-platform verification and support for more content types.