Optical-neural processor uses light diffraction to detect deepfakes at unprecedented speed and accuracy

UCLA researchers developed an optical-neural processor that uses light propagation to detect deepfake videos with nearly 98% accuracy while simultaneously analyzing multiple video streams. The system processes detection through physical light diffraction rather than conventional digital hardware, enabling parallel analysis of 15 or more videos in a single optical pass. The technology consumes minimal energy and is resistant to deliberate manipulation attempts, making it suitable for large-scale screening of AI-generated content.

The system combines a lightweight digital preprocessing stage with an optical analysis layer. Video data is first encoded into compact spatial, spectral, and temporal features, then converted into optical phase patterns displayed on a programmable light modulator. This encoded light travels through a passive optical decoder where detectors measure the resulting signal to produce authenticity scores—a design that fundamentally shifts computational burden from electronics to physics.

The technology demonstrates particular strength in sensitivity metrics, correctly identifying manipulated videos 99.86% of the time during testing. This low false-negative rate addresses a critical operational need: screening systems must catch fake content rather than mistakenly approve it. The ability to simultaneously process 15 or more videos in a single optical pass represents a substantial efficiency gain compared to sequential digital processing.