Does an image’s provenance prove that it is true?

“Is it real?” can mean several things

Someone sends you a photograph of a bridge and asks, “Is it real?” Do they mean that a camera captured it, that nobody edited it, that it shows the place named in the caption or that the event happened this morning? One short question contains several different checks.

Content Credentials provide a way to describe the provenance of digital images, video and other media. The C2PA explainer distinguishes verification of that record from a judgement that the depicted content is true.

This background article uses the public version 2.4 documents checked on 6 November 2026. It does not authenticate a particular post, and this page is not a file-signature verification service. The fictional cases above help us decide which question a piece of evidence can answer.

Separate the record, the signature and the binding

Three pieces help explain the system: statements about creation or editing, a digital signature over the relevant record, and information that binds the record to the asset. C2PA combines these pieces rather than relying on a label pasted beside a picture.

Validation can establish a relationship with a signer and check the integrity of the associated record. A signer is not automatically the real name of the person who pressed the shutter. Information about a tool or device and information about a human creator answer different questions.

Likewise, seeing a symbol on a page is not the same as validating the underlying record. A useful reading process asks what was recorded, which signer is involved and what validation result the viewer reports. A decorative badge alone cannot do that work.

An AI-generated picture can have valid provenance

Imagine an AI-generated blue bird whose creation method is recorded. Validating the record and its signature would not establish that the bird stood in that location. It could instead provide a verifiable connection to a statement that the image was generated.

The technical specification includes examples of recording generative-AI creation. A valid provenance record and an explicit AI origin are compatible. Content Credentials are not a pass mark reserved for camera photographs.

The reverse inference also fails: not seeing an AI-use statement does not by itself prove that no AI was involved. First ask where the available history begins and which steps it covers. A gap in evidence remains a gap; it does not turn into a guarantee simply because a reassuring interpretation would be convenient.

A caption can change the claim without changing the file

In another invented case, a photograph taken ten years ago is reposted, unchanged, with the caption “the bridge that opened today”. A verifiable history for the image would not automatically make the new caption accurate.

The reader now needs to compare the image’s context with the place and date asserted by the post. An old, genuine photograph can be used as misleading evidence for a new event. The image file and the claim made around it are separate objects of inquiry.

This is not a reason to discard provenance. It is a reason to give the tool a specific job. A record of origin can help answer questions about origin. Corroboration of an event addresses a different question. Keeping those jobs distinct makes the next check clearer.

Missing credentials do not mean a fake image

If no record can be found, several explanations remain possible. The creator may not have used a compatible tool; information may not have survived sharing; the available history may be incomplete. The absence alone cannot choose between these explanations.

C2PA’s harms document explicitly cautions against treating media without Content Credentials as automatically untrustworthy. Creators can have legitimate reasons not to use compatible tools.

Distinguish no available record, an available record that cannot be validated, and a record that validates. Even a validation problem is not, by itself, a finding that the depicted event was fabricated. Read what the verification result actually says and return to the original publisher or other evidence for the unresolved question.

Our view: build a checking sequence, not a single pass mark

Our editorial view is that provenance can add useful evidence to the way people read images. We would not teach readers that a badge means the surrounding story is true.

A practical sequence has three questions. Who is making which claim? What verifiable history is available for the image? Does the image actually support the claimed place, time and event? Completing one check should not silently stand in for the other two.

The “AI-generated illustration” captions on this library’s opening artwork explain the images’ purpose and production. They do not mean that this website issues or validates C2PA signatures. Our aim is for readers to enjoy imaginative artwork while recognising which materials are being offered as evidence.

An editorial perspective This section offers comparisons and questions based on facts in the sources. It does not claim a shared underlying principle or present new research results.