What Is C2PA? Content Credentials Explained (The "Nutrition Label" for Media)

By Calabi Labs Editorial Team · 2026-08-05

C2PA (Coalition for Content Provenance and Authenticity) is an open technical standard for attaching a tamper-evident, cryptographically signed record of where a piece of media came from and how it was edited. When you see a "Content Credentials" badge, you're seeing a C2PA manifest — a signed history stored inside the file itself, in a container format called JUMBF. It's designed to answer one question reliably: is this real, and who made it?

You can think of C2PA as a nutrition label for a photo or video. A nutrition label doesn't stop you from eating junk food; it just makes the ingredients legible. C2PA does the same for media: it doesn't decide whether content is good or bad, it makes the provenance — the origin and edit history — legible and verifiable. That's a bigger deal than it sounds, because for most of the internet's life there has been no trustworthy way to know where an image actually came from.

History: from the CAI to C2PA

The story starts in 2019, when Adobe launched the Content Authenticity Initiative (CAI) alongside The New York Times and Twitter. The goal was provenance: attach durable attribution and edit history to digital media so audiences could trace it back to a source. The CAI was the advocacy and adoption arm — it grew to thousands of member organizations.

In February 2021, the technical standards work consolidated into the Coalition for Content Provenance and Authenticity (C2PA), a joint project founded by Adobe, Arm, the BBC, Intel, Microsoft, and Truepic under the Linux Foundation's Joint Development Foundation. The C2PA's job was to publish the actual open specification, while "Content Credentials" became the consumer-facing brand for the badge and the little "cr" icon you now see on some images. You can read the specification and see member organizations at c2pa.org and contentcredentials.org. The timing was deliberate: generative AI was accelerating, and the coalition wanted a provenance standard in place before synthetic media became indistinguishable at a glance.

How the signed manifest works

C2PA's design rests on cryptographic signing rather than trust-me claims. Here is the mechanism, stage by stage.

1. An assertion is made. When a supporting tool creates or edits a file, it records structured statements — assertions — about what happened. These can include the capture device, the software used, edit actions (crop, color, generative fill), and a DigitalSourceType field. For AI-generated media, that field carries a value like trainedAlgorithmicMedia, which is a machine-readable declaration that a model produced the content.

2. Assertions are bundled into a claim. The tool gathers the assertions and hashes of the actual media data into a "claim." The hash binding matters: it ties the manifest to this specific pixel data, so if the image is altered, the hash no longer matches and the credential shows as broken.

3. The claim is cryptographically signed. The claim is signed with a certificate issued to the producing tool or organization. This signature is what makes C2PA tamper-evident: anyone can verify the signature against the signer's certificate, and any modification to the claim or the media invalidates it. This is the core reason C2PA is a strong provenance signal — it's verifiable math, not a self-reported label.

4. The manifest is stored in the file as JUMBF. All of this — assertions, claim, signature — is packaged into a C2PA manifest and embedded in the file using JUMBF (JPEG Universal Metadata Box Format, an ISO standard box structure). A single file can accumulate multiple manifests as it passes through tools, forming a chain of edits. In forensic terms, one AI-generated image can carry many JUMBF atoms and internal C2PA references documenting each step.

5. A verifier reads it back. A viewer, platform, or a tool like the Content Credentials inspector reads the manifest, checks the signatures and hashes, and displays the history. If everything validates, you get a trustworthy account of origin.

Who uses it, and why it matters

Adoption has spread across the media supply chain. Camera makers (Leica, Nikon, Sony) have shipped or piloted C2PA capture; Adobe embeds Content Credentials in Photoshop and Firefly; OpenAI attaches C2PA metadata to DALL·E and Sora outputs; and platforms including LinkedIn and others have begun surfacing the badge. News organizations use it to prove a photo wasn't manipulated. The common thread: C2PA is becoming the default way that "made by AI" gets declared in a machine-readable, hard-to-forge form.

What this means for a creator

The practical reality is that if you produce media with a C2PA-supporting generator, your file likely leaves the tool carrying a signed manifest that explicitly states it's algorithmic — stored as JUMBF atoms with internal references. That's excellent for transparency and provenance, which is the standard's whole point. It also means the manifest travels with your file unless it's deliberately removed, and that platforms can read it. Understanding C2PA lets you make an informed choice about provenance rather than being surprised by it. Note that C2PA is separate from invisible pixel watermarks; the manifest lives in the metadata layer, not the content itself.

Calabi Sanitizer can detect and strip leftover C2PA manifests (turning, in verified tests, 18 JUMBF atoms and 16 C2PA references down to 0) and shows you an ExifTool proof card confirming the manifest is gone.

FAQ

Is C2PA the same as Content Credentials?

They're two names for the same ecosystem. C2PA is the technical standard and the coalition that maintains it; "Content Credentials" is the consumer-facing brand for the badge and the signed manifest that the standard produces.

Where is a C2PA manifest actually stored?

Inside the file's metadata, in a container called JUMBF (an ISO box format). A file can hold multiple manifests forming an edit chain, each cryptographically signed and hash-bound to the media data.

Can a C2PA credential be removed?

Yes. Because it lives in the metadata layer, a manifest can be stripped from a file, and doing so removes the provenance record. It cannot be silently forged, though — altering a signed manifest breaks its signature and shows the credential as invalid.

Calabi Sanitizer automates the file-level cleanup described here — try it free at calabilabs.com (10 cleans, no card).