fightfake.ai
fightfake.ai builds open software so anyone can check that a video or photo was signed in a camera’s secure hardware, and that later edits were only the ones claimed, not a deepfake.
Example
Two clips look almost identical. The left is an authentic recording; the right was altered with an AI face swap. Looking at the faked video alone, it might be difficult to tell whether it is authentic, or what edits have been made.
Original
Fake
One control plays and scrubs both clips together.
Cryptography gives a test that looking cannot: authentic media can carry a signature made inside the camera when it was recorded, and a proof for every edit made afterwards. A deepfake can show neither. So the face swap above fails the check, while an honest edit, like the face redaction in the Example, passes, because it comes with a proof.
Three pillars make that possible: sign at capture, prove what changed, and check for yourself.
On the camera side, that pipeline needs hardware that binds the signature to the real pixel stream, not to bytes an app can invent. See Hardware for what is missing and how to work with us on it.
Notes on provenance, proof systems, and design.
Europe labels AI media, but deeper actions are needed.
Where in the camera should the hash be computed, and what should be hashed.
Do we really need a guarantee that photos and videos have not been manipulated before being published?