What Is a C2PA Content Credential? A Plain-English Guide

A C2PA Content Credential is a signed manifest embedded inside an image file. It records who made the image, what tool made it, and which edits happened since. The originating software writes it — not a human reviewer, not a watermark service — and the manifest travels with the file wherever the file goes.

Where C2PA lives in a file

Content Credentials are stored as JUMBF boxes, a generic container defined by the ISOBMFF family (the same spec family that powers MP4 and HEIC). In a JPEG, the JUMBF box sits inside an APP11 marker. In a PNG, it lives as its own chunk type. Nothing in the spec demands the credential be at a specific byte offset, so a scanner walks the file segment by segment to find it.

The credential is also hash-linked. Every edit signs the hash of the previous version, so any tampering breaks the chain. A complete scanner reads the entire history — first export, every Photoshop pass, every resave — without touching the pixel data.

What a credential actually contains

A typical manifest has four sections:

• Producer — the originating software and version, e.g. Adobe Firefly 3.0.
• Assertion — claims about the asset: c2pa.created, c2pa.ingredients, c2pa.actions ("trainedAlgorithmicMedia", "stitching", "colorAdjustment").
• Signature — an X.509 certificate pinning the issuer. Adobe's certificate chain, in practice.
• Hash chain — links every prior version so edits are auditable.

The hash chain is the part that makes the credential tamper-evident. If someone swaps a single pixel and resaves, the new hash no longer matches the asserted hash, and any C2PA-aware viewer will mark the file as "modified since signing".

Who uses C2PA today

Five years after the spec launched (the Coalition for Content Provenance and Authenticity was announced in 2021 by Adobe, Microsoft, BBC, Truepic, and others), adoption looks like this:

• Adobe Firefly — every export is signed, including derivatives from Photoshop and Illustrator.
• Microsoft Bing Image Creator / Copilot Designer — signed since late 2023.
• Leica cameras — first camera vendor to ship a hardware-signed credential.
• Sony cameras — second major camera brand, integrated mid-2024.
• Most social platforms — strip credentials on upload, mostly because they lack the rendering infrastructure.

How to read a credential yourself

Three options, in order of convenience:

1. Adobe's Content Authenticity Tool — a free browser extension focused only on C2PA.
2. ExifTool — exiftool image.jpg prints everything, including the JUMBF box, in raw form.
3. AI Metadata Ninja's tool — drag the file in and it lists the credential, the issuer, and every asserted action in plain English, alongside whatever XMP and EXIF the file also carries.

Why some platforms strip the credential

Most platforms aren't set up to render or verify Content Credentials, so they treat the JUMBF box like any other unknown metadata block and discard it. This is independent of the platform's broader policy on AI. Even platforms that welcome AI art tend to drop the trail by default because surfacing it needs extra infrastructure: signature validation, an X.509 trust store, and a UI element to render the credential badge.

What a credential does NOT prove

A Content Credential is not the same as a watermark, fingerprint, or AI-detection score.

• It does not prove the image is real. It proves the file came from a specific tool at a specific time.
• A human painter can use Firefly and get a perfectly valid credential.
• A bad actor can strip the credential and post the underlying pixels anywhere.

Think of it as provenance, not authenticity. The credential answers "where did this file come from" — not "is this picture true".