running a small independent publication. we get reader-submitted photos for local news pieces. need a verifiable process for confirming an image isnt AI generated before we publish. what’s the actual practice at bigger newsrooms
layered approach is the standard now. 1) check for embedded C2PA credentials in the file, that’s free and instant. 2) run through 2 independent image AI detectors (Hive, Optic, or similar) and look for agreement. 3) reverse image search to catch reuse. 4) if anything’s ambiguous, ask the submitter for the original RAW or out-of-camera JPEG with metadata intact. that last step alone resolves 80% of ambiguity because most AI tools can’t fake plausible EXIF and RAW data.
Plus one to Marc’s layered approach. our newsroom adds a 5th step: a written attestation from the submitter, kept on file. doesn’t catch a determined fraudster but creates legal exposure for the submitter that almost always surfaces honest answers in pre-publication conversations.
@Marc_Delrieu RAW request is a great default. people willing to submit RAW are essentially never submitting AI. thanks
the C2PA check is underused. its free, takes 5 seconds, and an increasing number of cameras and phones now embed it natively
Academic publishing is converging on a similar layered approach. We’ve added a ‘source artifact required’ line in our peer review guidance for photo evidence. The friction is up front, but appeals downstream have dropped sharply.