AiMetaOff / GUIDES

Understand Content Credentials and C2PA

Learn what C2PA Content Credentials can reveal, how signature and file-binding checks differ, and how to inspect embedded credentials in an image.

Content Credentials are tamper-evident records about how a digital file was made or edited. A credential can identify the software that exported an image, describe an edit, or record that an AI tool generated it. The Coalition for Content Provenance and Authenticity (C2PA) develops the open standard that lets tools create, read and check these records. A camera photo and an AI-generated image can both carry Content Credentials. C2PA Explainer

How do Content Credentials work?

One Content Credential is a C2PA manifest. A file can have a store containing multiple manifests, including records associated with earlier versions or source assets used to make it.

Assertions are individual records. They can describe actions, software, a capture device or ingredients such as an image used in a composite. The fields you see depend on what the signer recorded.
The claim references the assertions by cryptographic hash. A hash acts like a digest of the data, so a validator can check whether a referenced assertion still matches.
The claim signature is the digital signature over the claim. A validator checks it with the public key associated with the signing credential.
The content binding connects the manifest to the file. A hard binding uses hashes of the portions covered by the manifest. Those hashes let a validator check whether the credential matches those portions of this copy.

The active manifest is the record the validator uses to check the current asset. Earlier manifests and ingredient records can provide context about what went into it. C2PA Technical Specification

What does a verified result tell you?

Start by checking whether a supported container was found and whether the manifest could be read. Then look at the individual results.

Claim signature. Did the signature verify against the signed claim and the signing key?
Assertion hashes. Do the checked assertions still match their references in the claim?
File binding. Do the covered parts of this file still match? Changes to those parts can cause a mismatch while the claim signature still verifies.
Certificate validity. Does the available evidence support certificate validity? The offline result does not include a live revocation check.
Signer trust. Does the signing certificate chain lead to a trust anchor accepted by the validator? The signer might be the software vendor or publisher. A name shown in a certificate field alone does not establish who took the photo.

A passed check supports a specific conclusion. A failed check signals a problem in that check. An unavailable check leaves the question open. The standard distinguishes a well-formed, valid or trusted manifest from a valid asset, which additionally requires verification of the covered content. C2PA Technical Specification, validation states

Are Content Credentials proof that an image is authentic?

A matching binding is evidence that the covered parts of the file match the signed version. Even a trusted signer can sign a staged photo, an edited image or an incomplete account of how it was made. Assess the recorded statements and the image's context separately. For example, a genuine signed photo can still be shared with a misleading caption. C2PA Explainer

Why can signature and file binding disagree?

In an AiMetaOff test using a signed JPEG, the claim signature and file binding both verified on the original sample. After the test changed image data covered by the binding, the signature still verified and the file binding failed. The signed claim had survived, but it no longer matched the checked image data.

How to inspect embedded Content Credentials

Start with the original download if you have it. To understand what a recipient received, check that exact copy as well.
Open Inspect File and choose the file. AiMetaOff processes it in your browser and automatically checks for supported embedded credentials.
Read the recorded software and the separate signature, assertion-hash and file-binding results. Open View technical details for available actions, declared source types, certificate fields and validation findings.
Check the scope of the result. AiMetaOff reports signer trust as unverified. It does not fetch external manifests or perform live certificate revocation checks. If a check is unavailable, keep that uncertainty when describing the file.

Inspect Content Credentials in your file

Embedded, external and durable credentials

An embedded manifest store travels inside the file. A store can also be supplied separately, for example as a sidecar file, or referenced at an external location. A durable credential can be discovered in a compatible repository through a soft binding such as a fingerprint or invisible watermark. Discovery can reconnect an image to a record. Whether that record's hard binding matches this exact copy is a separate check. C2PA Technical Specification

A C2PA-aware editor can sign a new version and reference earlier provenance. A crop, re-encode or conversion that changes covered data can break the old binding. A screenshot or platform export can produce a new copy without the original embedded store. If someone shares a platform copy, inspect it and the original separately. C2PA Explainer · OpenAI provenance documentation

AiMetaOff checks supported embedded records offline. It does not open separate sidecars or recover a durable credential from a watermark.

Content Credentials on AI-generated images

A manifest can record a creation or editing action along with a declared digital source type. For example, the C2PA standard provides trainedAlgorithmicMedia for media generated by a trained algorithm. An image with AI-assisted edits may carry a different source declaration. Interpret the action, source type and signer together. C2PA Technical Specification

Adobe documents that assets created entirely by Firefly, such as Text to Image outputs, receive Content Credentials with details such as the issuer, tool and action. Some credentials may also be stored in Adobe's public credentials cloud. See the Adobe Firefly export guide for an example of those fields. Adobe Firefly Help

Frequently asked questions

What is the difference between C2PA and EXIF or XMP?

EXIF commonly stores camera details such as exposure, capture time and orientation. XMP can carry descriptive and rights information, among other fields. Those ordinary metadata fields do not by themselves provide a signed claim bound to the image. A file can contain all three, and XMP can also carry a reference to an external C2PA manifest. CIPA standards · Adobe XMP specifications · C2PA Technical Specification

Does “not found” mean an image was made by a human?

No. The inspected copy may never have had a credential, or it may have lost it during export or sharing. An associated record may exist elsewhere. An unsupported file or an unavailable result also leaves the image's origin open.

Can Content Credentials be removed or recovered?

The C2PA Remover can remove supported embedded stores in eligible JPEG, PNG and WebP files. AiMetaOff writes a separate copy, re-inspects it and checks that the compressed image data was preserved without re-encoding. External records, watermarks and platform labels can persist. Recovery depends on an accessible record and a compatible way to find it. Adobe's Inspect documentation