Researchers at the University of California, Los Angeles (UCLA) have developed a light-powered artificial intelligence system that can detect deepfake videos with nearly 98% accuracy while screening up to 15 videos simultaneously.

Unlike conventional digital deepfake detectors, which typically process videos sequentially, the new optical-neural processor uses the physical propagation of light to analyse multiple video streams in parallel. The approach could reduce processing time and energy consumption, according to the researchers.

The technology, detailed in a study published in eLight, was developed by a UCLA team led by Professor Aydogan Ozcan as a high-throughput first line of defence against manipulated and AI-generated videos.

In tests involving 15 Celeb-DF videos, the processor achieved 97.79% overall accuracy, with 99.86% sensitivity and 95.72% specificity. When expanded to process 18 videos simultaneously, its accuracy remained at 96.13%.

The researchers further improved the system by adding two passive diffractive optical layers. In tests involving more challenging video manipulations, the additional layers increased detection accuracy by about 6.8% without significantly increasing energy use or processing time.

The system was also tested on videos generated by Google’s VEO-3 model, which can produce highly realistic synthetic footage. With minimal fine-tuning, the processor achieved 94.80% accuracy and 97.61% sensitivity when analysing previously unseen VEO-3 videos.

According to the researchers, the optical design is also more resistant to certain adversarial attacks because key computational parameters are physically embedded in the hardware, making them more difficult to reproduce or reverse-engineer.

The processor continued to function despite image noise, blur, JPEG compression and minor optical misalignments.

The team said the technology is not intended to replace conventional digital detectors. Instead, it could work as a rapid screening layer, filtering large volumes of video and flagging suspicious content for more sophisticated digital systems to analyse.

The researchers said such a hybrid approach could eventually be used for large-scale content moderation, media authentication, surveillance and other security applications.

The study was co-authored by Parnian Ghapandar Kashani, Dr Shiqi Chen and Professor Aydogan Ozcan of UCLA’s departments of Electrical and Computer Engineering and Bioengineering, and the California NanoSystems Institute.