Spotting an AI-generated image is getting harder by the day. OpenAI wants to change that, and it is using Google as a partner to accomplish this behemoth task.
OpenAI has announced a significant update to how it handles content provenance, which is just a fancy way of saying how you can tell where an image came from and whether it was AI-generated.
The update has three key parts: C2PA conformance, a new watermarking partnership with Google, and a public tool you can use to verify images yourself.
So what is actually changing?
OpenAI has been adding metadata, called Content Credentials, to images generated by its tools since 2024. This metadata is embedded in all generated images and tells you how it was created. The company has now become a C2PA Conforming Generator, which means other platforms can reliably read and find the information when they encounter OpenAI-generated content.
The bigger news is the partnership with Google. OpenAI is now adding Google DeepMind’s SynthID watermarking to images generated through ChatGPT and the OpenAI API. Unlike metadata, which can be stripped when you screenshot or re-upload an image, SynthID embeds an invisible watermark directly into the image. It is designed to survive the transformations that can typically erase metadata-based detection.
The two systems work together. The metadata carries detailed information about the image’s origin, while the watermark acts as a backup when the metadata doesn’t make it.
How can you check if an image is AI-generated?
OpenAI is previewing a public verification tool at openai.com/verify. You can upload an image, and the tool will check for both Content Credentials and SynthID watermarks to determine if it was generated using OpenAI’s tools.
One important caveat: if the tool finds no watermark or metadata, it will not definitively say the image is not AI-generated. This is because the watermark and metadata can still be spoofed.
Also, not all generative AI companies have partnered up with Google to include SynthID in their images. Unless SynthID or something similar is universally adopted by every generative AI model, no verification tool can tell you with 100% certainty that an image is not generated by AI.
Still, it’s a step in the right direction. AI-generated images have become a mayhem and are widely used for nefarious purposes. A system to recognize ones is absolutely needed, and I am happy Google and OpenAI are paving the way for it.