Reality is losing the deepfake war

February 5, 2026·49 min·4 clips

OpenAI admits its own C2PA metadata is incredibly easy to strip, exposing a fatal flaw in the fight against deepfakes.

The question starts with a photo. Patel uses that frame to ask whether images and videos can carry labels strong enough to hold up public trust while generative AI changes creative tools. This is not just a Photoshop and Canva episode, though both sit inside Jess Weatherbed's Verge beat. Weatherbed is there because the issue now runs through editing software, artists, creatives, and anyone trying to understand what they are looking at. C2PA is the standard on the table. Before the break, she has been explaining where the provenance effort came from and why AI image labels have moved so slowly. The hard part is not only creation. Phone makers, camera providers, and the rest of the photography ecosystem all have to make authenticity data work. Then distribution gets in the way. Social platforms need to preserve and show that metadata in a way ordinary viewers can actually see. That is where the whole thing gets shaky. A label only matters if it survives the trip from camera or editing tool to platform to viewer. Patel keeps pulling the conversation back to the practical question: what counts as a photo when capture, editing, generation, and platform display are hard to separate? Weatherbed brings the standards context. Patel keeps pressing on how labeling, metadata, and visibility work after media leaves the device or app. There is ad copy around the interview, but the conversation itself is about the fragile infrastructure behind trust. The episode treats provenance as necessary plumbing, and plumbing breaks when the companies moving reality around do not coordinate. The closing lands cleanly. After all the metadata talk, Patel asks listeners what they think a photo is, which is the quiet point of the whole thing.

As heard by us

A clear, practical look at why labeling AI images gets harder once platforms control what users see.

Decoder frames the deepfake problem as a question of infrastructure: can photos and videos carry useful labels as they move through cameras, editing tools, phone makers, and social platforms?

Read the full review in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from Decoder with Nilay Patel