Provenance, C2PA and visible labels
Provenance checks look for direct context about a file's creation or editing history. IA Checker treats signed credentials, metadata and visible labels as distinct evidence sources.
C2PA content credentials
The image worker can use c2patool to validate a C2PA manifest and inspect its assertions. A valid signed claim can add direct creation or editing context to the report.
A missing C2PA claim is not evidence that an image is human-made. Credentials can be stripped by screenshots, exports, unsupported editors or distribution platforms.
Metadata and visible labels
EXIF and other embedded fields can contain software or workflow hints. Separately, visible-label OCR can read explicit marks such as an AI-generated label or a provider watermark rendered into the pixels.
SynthID and OpenAI source signals
IA Checker does not currently claim automatic SynthID verification because no supported public verification adapter is wired into the production pipeline. The interface can report only what the available check actually establishes.
OpenAI-related provenance is read through validated C2PA claims when present. There is no separate public OpenAI Verify API in the current workflow.