OpenAI has launched a research preview of a tool that verifies uploaded images for signs that they may have been generated with OpenAI tools. The page says the system checks for provenance signals linked to OpenAI-generated images and identifies whether it detects C2PA Content Credentials, SynthID watermarks or no supported signal.
According to OpenAI, the verifier is designed to detect images created with ChatGPT, the OpenAI API or Codex. The company says users should upload one image at a time and, for best results, crop screenshots tightly around the image rather than uploading files containing multiple images.
OpenAI says provenance matters because it can help people better understand where digital content came from and whether artificial intelligence played a role in creating or editing it. The company says such context may reduce confusion and deception online and provide more clarity for journalists, platforms, creators, researchers and the public.
The tool relies on two signal types: C2PA Content Credentials and SynthID. OpenAI says Content Credentials use metadata that can be removed, while SynthID is an invisible watermark embedded into the image itself and is designed to withstand changes such as cropping, filters and lossy compression.
OpenAI also says the tool cannot determine whether an image is accurate or misleading. A detected signal only suggests the image likely originated from OpenAI tools, and a missing signal does not prove the image was not made by OpenAI, since metadata may have been stripped, watermarks may be degraded, or the image may have been created using older models or before provenance signals were available.
The company says uploaded images are processed only for supported provenance signals, are not stored unless legally required, and are not used to train its models. Readers interested in testing the system can access the new verification platform here: OpenAI Verify
I truly appreciate you spending your valuable time here. To help make this blog the best it can be, I would love your feedback on this post. Let me know in the comments: How could this article be better? Was it clear? Did it have the right amount of detail? Did you notice any errors?
If you found any of the articles helpful, please consider sharing it.