Hany Farid Says AI Deepfakes Have Defeated His Own Eyes

Hany Farid spent more than two decades as the person governments, law enforcement agencies, journalists, and human rights organizations called when they needed to know whether a video or photograph was real. According to The New York Times, the 60-year-old UC Berkeley professor had already shown through his own research that most people can no longer distinguish a real photograph from a digital creation, a real voice from an AI clone, or a real video clip from a fabrication. In the last six months, Farid has been failing those same tests himself.

He described the experience in language blunt even by the standards of someone who built a career in forensics: "Every image I see, I'm drawing lines for shadows and doing geometry in my head, trying to figure out what I'm looking at. It's over. Within a year or two, our whole visual system will be utterly useless." Elsewhere he put it more simply: "I feel like I am going blind." The backdrop the article sets is concrete: cybersecurity firm DeepStrike reportedly found that deepfake content grew roughly 900% over the past year, though take that figure as a claimed data point rather than a settled count.

Farid is departing UC Berkeley effective June 30, 2026, returning to Dartmouth College as of July 1. He and his wife have reportedly begun making plans to leave California and trade Silicon Valley's tech culture for a farm in rural Vermont.

What the reporting does not give you is what Farid thinks should replace visual inspection -- whether cryptographic content authentication, platform-level provenance tools, or something else can fill the gap. The question matters practically: organizations that built verification workflows around human review of video evidence, from newsrooms to courtrooms to humanitarian investigators, now have the field's most credentialed detector on record saying that approach is no longer reliable.