A new authentication method uses the physics of light itself to tell real video from AI-generated footage.
Researchers have proposed exploiting the Moire effect — interference fringes that appear when a camera films a two-layer grating structure — as a built-in signature for genuine footage. They derived what they call the Moire motion invariant: fringe phase and grating displacement stay linearly coupled through optical geometry, regardless of viewing distance. A verifier checks whether that correlation holds. Testing across multiple state-of-the-art video generators, real footage consistently produced a distinct correlation signature that AI-generated clips did not.
Most deepfake detection tools chase artifacts — compression glitches, unnatural blinking, inconsistent lighting — and labs race to patch those tells with each new model release. This approach flips the problem: instead of looking for what AI gets wrong, it anchors verification to something AI cannot simulate without physically recreating camera optics. That makes it structurally harder to defeat by simply training a better generator.
The method requires a physical grating in the scene, which limits its use to pre-planned, controlled recordings — courtroom footage or broadcast news, perhaps, but not your phone video after the fact. Still, for high-stakes verification contexts, a forgery-proof optical signature beats another ML classifier in an arms race.