Guide

What is C2PA? Content Credentials, explained

C2PA is an open standard for attaching a signed, tamper-evident record of how a photo, video or sound file was made. Here is how it works, who uses it and what it can and can’t tell you.

The short version

A Content Credential is like a nutrition label for a media file. It lists who made the file, with what device or software, and what was done to it afterwards — and it is cryptographically signed, so any change to the file or the label afterwards is detectable. When a generative AI tool creates an image, the credential says so in a standard, machine-readable way.

How it works

The manifest

Each time a C2PA-aware tool creates or edits a file, it writes a manifest. A manifest contains assertions — statements such as “this file was created”, “these edits were made”, “this is a thumbnail of the result”, “this is a hash of the image data” — and a claim that lists the assertions and names the software. The claim is then signed with the tool maker’s certificate.

Binding to the content

One assertion is a hash of the file’s actual content. If anyone changes a single pixel without writing a new manifest, the hash no longer matches and a validator reports the file as tampered.

The chain of edits

When a credentialed file is edited, the new manifest includes the old one as an ingredient. The result is a history: captured on a camera, cropped in an editor, a background extended with generative fill, exported for the web.

Saying “AI” in a standard way

Actions carry an IPTC digital source type. The value trainedAlgorithmicMedia means the content was created by a generative model; compositeWithTrainedAlgorithmicMedia means real content combined with generated parts; digitalCapture means a camera captured it. This is what lets a detector say “made with AI” with certainty instead of guessing.

Why most files don’t have one

Credentials are metadata, and the internet is very good at destroying metadata. Screenshots never carry it. Most social networks, messaging apps and image hosts re-encode uploads and drop it. Older editing software discards what it doesn’t understand. On top of that, most cameras and phones in use today don’t create credentials at all. So in practice, credentials are excellent evidence when present and silent when absent — which is why GPTTrace pairs them with metadata and content analysis. To inspect a file’s manifest, use the C2PA viewer.

Frequently asked questions

What does C2PA stand for?

The Coalition for Content Provenance and Authenticity, a standards body formed in 2021 by Adobe, Microsoft, the BBC, Intel, Arm and Truepic, joined later by Google, OpenAI, Meta, Amazon, Sony and many others. “Content Credentials” is the consumer-facing name for its manifests.

Is C2PA a watermark?

No. A watermark is hidden in the pixels or audio samples; a C2PA manifest is signed metadata stored alongside them. Manifests carry rich, verifiable detail but can be stripped. Watermarks such as SynthID survive editing but carry little information and need the vendor to read them. The two are complementary, and the C2PA specification can reference watermarks.

Do social networks keep Content Credentials?

Mostly not in the file you download. Some platforms read credentials on upload and show a label such as “AI info” or a “CR” icon, but serve a re-encoded copy without the manifest.

Can C2PA prove an image is real?

It can show that a trusted device or tool signed the file and that the pixels have not changed since. It can’t prove the scene wasn’t staged, and a missing credential proves nothing at all.

How do I add Content Credentials to my own work?

Use software or hardware that supports them: recent Adobe apps, some Leica, Sony, Nikon and Canon cameras, Google Pixel phones, and various newsroom and capture apps. Keep credentials enabled when exporting and share original files where you can.