UCLA researchers use light to spot deepfake videos

Researchers led by Aydogan Ozcan at UCLA have built a deepfake detector that does part of its computing with light instead of electronics. The work was published in the journal eLight and reported by Optica's Optics & Photonics News on October 5, 2026.

How it works

• Small neural networks first pick out clues from a few frames of each video.
• Those clues are written into a pattern of light, so one beam carries information from many videos at once.
• Light detectors read the result: the difference between each pair of detectors says whether a video is real or fake.

The results

On the Celeb-DF benchmark, the system was 97.8% accurate while checking 15 videos at the same time, and 96.1% with 18 videos. The authors say it uses much less energy than the digital detectors they compared it with.

Why it matters

Platforms need fast, cheap ways to screen huge amounts of video. This is still a lab prototype, but it shows how new hardware could help in the fight against deepfakes and misinformation.