OpenAI launches a tool to verify AI-generated images and audio
Checks uploads for C2PA metadata and SynthID watermarks from ChatGPT, the API, or Codex, with results available via the API too.
OpenAI has published a public verification tool for checking whether an image was made with its products. Uploaded files are checked for two provenance signals: C2PA Content Credentials, which are metadata attached to the file and can be stripped, and SynthID, an invisible watermark embedded in the content itself that is designed to survive cropping, filters and lossy compression.
The tool covers images and audio generated with ChatGPT, the OpenAI API, or Codex, including files OpenAI's models have edited. A detected signal means the content likely came from OpenAI tools, but it says nothing about whether the content was later modified, is accurate, or was created by another company's model. No signal found doesn't rule out OpenAI origin either: metadata may have been stripped, the watermark degraded, or the file predate provenance signals. For best results, OpenAI recommends uploading one file at a time, avoiding cropping or format conversion for images, and using audio clips between 10 and 60 seconds. Developers can also run verification through the OpenAI API.
