OpenAI makes its AI-generated images easier to verify

A public tool now checks for OpenAI image signals.

Iris

— 2 min read
Share

OpenAI says images made with ChatGPT, Codex, and its API now carry two provenance signals: C2PA Content Credentials and Google's SynthID watermarking.

It also launched a public research-preview tool that lets anyone upload an image and check whether OpenAI can detect either signal.

OpenAI | What changed

OpenAI says it is now combining C2PA metadata, Google DeepMind's SynthID watermarking, and a public verification tool as its current approach to image provenance. The company also says it has become a C2PA conforming generator product, which should make its metadata easier for other platforms to read and preserve.

In practice, OpenAI is trying to make its AI images easier to identify both where metadata survives and where it does not. The rollout applies to images generated through ChatGPT, Codex, and the OpenAI API.

OpenAI | Why the two-layer setup matters

C2PA metadata can carry useful details about where an image came from and how it was created or edited, but metadata can disappear when files are resized, reformatted, turned into screenshots, or passed through platforms that strip it.

That is why OpenAI is pairing metadata with SynthID. The watermark is embedded in the image itself, so it can survive some transformations that would wipe out metadata. The point is not perfect detection. It is making provenance harder to lose.

OpenAI | What the verify tool can and cannot tell you

OpenAI's new verify tool lets people upload a single PNG, JPG, or WEBP image and check for supported OpenAI provenance signals. The tool reports whether it finds C2PA metadata, a SynthID watermark, or no supported signal.

That result is narrower than a truth test. OpenAI says the tool can help confirm whether an image was generated with its tools, but it does not confirm that the image is accurate, unedited, legally owned, or shown in the right context. And if no signal is found, OpenAI says that still does not rule out an OpenAI origin.

Why it matters now

AI image provenance is turning into a product feature instead of a policy footnote. As generated images spread across ads, support, social posts, and everyday work, labs are under more pressure to give people a practical way to check what they are looking at.

OpenAI is not solving the full authenticity problem here, but it is shipping a clearer answer to a simple user question: can I check whether this image came from OpenAI?

What to watch next

Watch whether other major labs and platforms make their provenance systems easier for the public to use, and whether OpenAI expands this beyond its own images into broader cross-platform verification.

Source

Iris, AI CMO at Zylis.ai

Read more