Content Credentials (C2PA)
Turn on Content Credentials (C2PA) under Identity Document in the workflow builder. It checks uploaded files and document photos for C2PA Content Credentials: whether the image is unchanged since it was signed, who signed it, and whether it says it was made with AI or edited. It covers the front and back photos of the document as the browser sent them and, when the workflow asks for one, the uploaded proof of address (JPEG, PNG, WebP, HEIC, HEIF, AVIF; a PDF is not applicable). A live camera photo normally carries no credentials, which is expected and neutral; an injected image (a virtual camera or a scripted upload) may still carry an AI generator's credentials. The check runs on our own validator in the background and never delays the person; the image and its result are not kept there. Workflows that turned it on under Proof of Address keep it, and now cover the document photos too.
| Finding | Reason code | Default |
|---|---|---|
| The credentials say the image was generated or composed with AI | CONTENT_CREDENTIALS_AI_GENERATED | Decline (or review) |
| The image was changed after signing, or the signature is broken | CONTENT_CREDENTIALS_INVALID | Review |
| The credentials record content edits or a composite | CONTENT_CREDENTIALS_EDITED | Review |
| No credentials (most images), or the check could not run | none | Never held against the verification |
The signed result and the full report carry content_credentials.images: per image its source (document_front, document_back, proof_of_address), state, verdict (trusted, valid_untrusted, invalid, no_manifest), ai_generated, edited, camera_capture, digital_source_type, signer, claim_generator and failure_codes. A camera-signed photo is a good sign but never approves a verification on its own. The check is charged once per verification, however many images it checked, and only when it checked one; it is refunded when there was nothing to check or the validator was unavailable.