In a world where images, audio, and video can be perfectly faked, one of the most promising responses isn't detecting fakes — it's proving the origin of the real. That's the goal of content provenance standards like C2PA (the Coalition for Content Provenance and Authenticity).
The idea: a nutrition label for media
C2PA attaches tamper-evident metadata to a piece of content — a cryptographically signed record of where it came from and what happened to it. Was this photo taken by a camera, or generated by AI? Was it edited, cropped, color-corrected? The provenance data travels with the file and can be verified, like a chain of custody. It doesn't judge whether content is "true" — it records its history so you can decide.
Instead of asking "is this fake?", provenance answers "where did this come from, and what was done to it?" — a question you can actually verify.
How it works
At creation (a camera capture, an AI generation) the origin is recorded and cryptographically signed. Each edit adds a signed entry. Because it's signed, tampering breaks the seal — you can't quietly alter the content without invalidating the provenance. Viewers (or platforms) can check the record and see the trail.
Why it beats detection
Detection is reactive and loses to better generators. Provenance flips the burden: rather than trying to catch every increasingly-perfect fake, you establish trust for authentic content at the source. AI-generated content can also be labeled as such through the same system, making disclosure verifiable rather than optional.
The catch
Provenance only works if it's widely adopted — by cameras, editing tools, AI generators, and the platforms that display content — and if people actually check it. Metadata can also be stripped (though that itself is a signal). So C2PA isn't a complete solution, but it's a foundational piece of the answer to synthetic media: a shared, verifiable way to know where content came from, which detection alone can never provide.
