{"version":1,"type":"story","url":"https://digestai.news/story/ucla-builds-optical-ai-that-spots-deepfakes-with-nearly-98-accuracy","json":"https://digestai.news/story/ucla-builds-optical-ai-that-spots-deepfakes-with-nearly-98-accuracy.json","markdown":"https://digestai.news/story/ucla-builds-optical-ai-that-spots-deepfakes-with-nearly-98-accuracy.md","slug":"ucla-builds-optical-ai-that-spots-deepfakes-with-nearly-98-accuracy","headline":"UCLA builds optical AI that spots deepfakes with nearly 98% accuracy","summary":"UCLA researchers have created an optical‑neural processor that uses light to examine multiple video streams at once, achieving deep‑fake detection accuracy of 97.79% across 15 videos in a single optical pass. The system combines a lightweight digital encoder that extracts spatial, spectral and temporal features with a programmable spatial light modulator, then lets the encoded light propagate through a passive optical decoder that directly outputs authenticity scores.  \n\nThe prototype handled 15 Celeb‑DF videos simultaneously, delivering 99.86% sensitivity and 95.72% specificity, and maintained 96.13% accuracy when scaled to 18 videos. Adding two diffractive layers boosted performance by about 6.8% without extra power. Tests on Google’s VEO‑3 generated videos showed 94.80% accuracy after minimal fine‑tuning. The researchers say the hardware‑embedded computation makes the detector harder to fool with adversarial attacks and robust to noise, blur, compression and misalignments, positioning it as a high‑throughput first‑line defense for large‑scale content moderation.","keyPoints":["Optical processor analyzes 15 videos simultaneously, reaching 97.79% detection accuracy","Sensitivity 99.86% and specificity 95.72% give a false‑negative rate of ~0.14%","Adding two passive diffractive layers improves accuracy by ~6.8% without extra energy"],"whyItMatters":"A parallel, low‑energy detection layer could dramatically cut the compute and power needed for large‑scale deep‑fake screening, strengthening content moderation and security against evolving AI‑generated media.","category":{"slug":"research","name":"Research","url":"https://digestai.news/category/research"},"entities":{"companies":["UCLA","Google","Light Publishing Center","Changchun Institute of Optics","Fine Mechanics and Physics","CAS"],"models":["VEO-3"],"people":["Aydogan Ozcan","Parnian Ghapandar Kashani","Shiqi Chen"]},"firstPublishedAt":"2026-10-02T21:38:39Z","updatedAt":"2026-10-02T21:38:39Z","sourceCount":1,"hasPrimarySource":false,"sources":[{"outlet":"sciencedaily.com","title":"This light-powered AI can spot deepfakes with nearly 98% accuracy","url":"https://sciencedaily.com/releases/2026/09/260929053534.htm","publishedAt":"2026-10-02T21:38:39Z","type":"press","primary":false,"lead":true}],"sourceNotes":null,"discussions":[{"site":"Reddit","url":"https://www.reddit.com/r/artificial/comments/1ww5zwo/this_lightpowered_ai_can_spot_deepfakes_with/","points":null}],"thread":null,"cite":{"text":"Digest AI, \"UCLA builds optical AI that spots deepfakes with nearly 98% accuracy\", 2 October 2026, https://digestai.news/story/ucla-builds-optical-ai-that-spots-deepfakes-with-nearly-98-accuracy","publisher":"Digest AI","title":"UCLA builds optical AI that spots deepfakes with nearly 98% accuracy","datePublished":"2026-10-02T21:38:39Z","url":"https://digestai.news/story/ucla-builds-optical-ai-that-spots-deepfakes-with-nearly-98-accuracy"},"generatedBy":"Written by Digest AI's editorial model from the linked sources; the sources are the record.","license":"Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse"}