A signed manifest documents a claim, not a verified fact

The shot

A clip leaves a camera or a generative video tool and picks up a small signed record of where it came from. The Coalition for Content Provenance and Authenticity (C2PA) calls this record a manifest, and its public name is Content Credentials: a signed structure that can travel with a file through a color grade, a VFX composite, an export, and a platform upload, recording what changed and, where a tool supports it, whether generative AI was involved at a given step. As retrieved on 16 September 2026, C2PA's specification is at version 2.1, a living document that member companies implement at their own pace.

What the documents show

The C2PA technical specification defines a manifest as a digitally signed, tamper-evident structure that references assertions: statements about how an asset was created or edited, which ingredients (other assets) went into it, and which actions were taken, such as 'create,' 'embed,' or 'apply filter.' The specification is explicit about its own limits, stating that C2PA specifications should not pass value judgments about whether provenance data is 'good' or 'bad,' only whether the assertions inside a manifest can be validated as associated with the underlying asset. Trust signals, in the spec's own language, inform a viewer's judgment; they do not settle it. That is a standards body describing its own scope, not an independently verified claim about any particular video and not a vendor's promotional pitch. When a tool maker says its product 'supports Content Credentials,' that is the vendor's own claim about an implementation, separate from what the standard itself proves.

The workflow

A production that wants a credential trail needs every tool in the chain to cooperate. A camera or generation tool that supports C2PA signs the first manifest, and an editor's software that also supports it adds a signed assertion at each save or export. Because the binding is cryptographic, an edit made in a tool that ignores the standard breaks the chain, so a mixed pipeline of AI, color, and edit software shows a trail that stops at the first unsupported step. A platform or public inspection tool can then display whatever chain survived, but a viewer sees only what was signed, not everything done to the file.

What the tool does not change

A manifest does not adjudicate truth. The specification's own refusal to grade provenance data as 'good' or 'bad' means a human still decides whether a signer is trustworthy, whether a broken chain reflects tampering or an ordinary tool gap, and whether a claim recorded at capture was accurate. None of that review work moves into the standard itself.

• Does every tool in this production's pipeline actually implement C2PA, or does the chain break somewhere in the middle?
• If a platform strips metadata on upload, what happens to the credential a filmmaker attached?
• Is a vendor describing its own implementation, or is that claim independently checkable against the specification?

A signed manifest is a documented claim about a file's history, built to survive edits that respect the standard. It is not, by the specification's own account, proof that a video is authentic.