ExifZero What is C2PA (Content Credentials)?
By the ExifZero team · Updated
C2PA — the Coalition for Content Provenance and Authenticity — is an open standard for attaching a tamper-evident history to images, video, audio and documents. Products show it to people as Content Credentials, often with a small “cr” icon. The coalition was founded by Adobe, Microsoft, the BBC, Intel and others; OpenAI, Google, Meta, Sony, Leica, Nikon and Samsung have since joined or adopted it.
What is inside a C2PA manifest
A manifest is a small package of signed statements stored inside the file. It contains:
- A claim — which app or device made the manifest (the “claim generator”), plus the file's title and format.
- Assertions — individual statements. The most important are actions (created, opened, edited, cropped, color-adjusted…), ingredients (the source files used), a hash of the image data, a thumbnail and, optionally, author information and AI-training preferences.
- A signature — made with a certificate that names the signing organisation. Changing the image or the claim breaks the signature.
Every C2PA-aware edit adds a new manifest and keeps the previous ones, so a file can carry its whole history. In JPEG files the manifests live in APP11
segments, in PNG files in a caBX chunk and in WebP in a C2PA chunk, all using the JUMBF box format.
How C2PA labels AI-generated images
Each action can carry a digital source type taken from the IPTC vocabulary. Generators such as ChatGPT image generation and Adobe Firefly write
a c2pa.created action with the source type trainedAlgorithmicMedia. Editing a real photo with generative fill adds an action
with compositeWithTrainedAlgorithmicMedia. Viewers like LinkedIn and the Content Credentials verify site turn this into an “AI-generated”
label.
What C2PA can and cannot tell you
| It can show | It cannot show |
|---|---|
| Which organisation signed the history | Whether the scene in the picture is true |
| That the file was not changed after signing | The history of a file whose manifest was removed |
| Declared AI generation or AI edits | AI use by tools that do not write C2PA |
The biggest limitation is that metadata is easy to lose. Screenshots, many editing apps, messaging apps and most social networks drop the manifest. That is why some vendors also embed invisible watermarks in the pixels (Google's SynthID, for example) and link them to the credentials through a cloud lookup.
How to check an image
Drop it into our C2PA viewer. It shows the claim generator, actions, ingredients, signer, thumbnail and training preferences, entirely in your browser. For a trust decision that verifies the signature and certificate chain, use the coalition's own verification tool as well.
Privacy considerations
Credentials are designed to travel with a file, and that includes things you may not intend to publish: the names of source files, thumbnails of earlier versions, and identity assertions that can name a person or account. If you share files with credentials, look at what they contain first.