What Is C2PA Metadata?

The 60-second version

C2PA stands for the Coalition for Content Provenance and Authenticity. It is an open standard, developed by Adobe, Microsoft, the BBC, Intel, Truepic and OpenAI, that embeds a tamper-evident cryptographic record into a media file. Think of it as a digital chain of custody — every editor, every camera, every AI model that touches the file can append a signed claim.

For ChatGPT images, the relevant claim says: "this image was generated by OpenAI's gpt-image-1 model at timestamp X, signed with our certificate."

How it's stored inside the file

C2PA claims are packaged inside a JPEG/PNG/WebP using a container format called JUMBF (JPEG Universal Metadata Box Format). The magic bytes6A 75 6D 62 mark the start of a JUMBF box, and our detector simply looks for those bytes plus a handful of ASCII identifiers like c2pa andcontentauthenticity.

What's actually inside a ChatGPT C2PA manifest

• claim_generator — which OpenAI model produced the image
• created — ISO 8601 generation timestamp
• actions[] — list of generation steps
• signature_info — issuer (OpenAI), signing cert, algorithm
• hash — SHA-256 of the image bytes
• thumbnail — a small preview embedded inside the file

Who reads it?

Adobe Photoshop, Microsoft Edge, social platforms (Meta is rolling it out), stock photo sites, journalism workflows, and any tool that integrates the C2PA SDK. Once they read the manifest they can show a "Made with AI" badge — or refuse the upload entirely.

How to remove it

Use our free ChatGPT image watermark remover. It runs entirely in your browser, re-encodes the image through the HTML5 Canvas API, and produces a clean file with no JUMBF box. Verify the result with the verify tool.