Hany Farid¶
Hany Farid is a professor at UC Berkeley's School of Information and co-founder and chief science officer at GetReal Security. Over two decades, he has been a leading researcher in digital forensics, media authentication, and the detection of manipulated images, video, and audio. His work addresses the detection and forensic analysis of AI-generated and synthetic media in contexts spanning journalism, law enforcement, and national security.
Sources in this wiki¶
- Academic Review: Political Deepfakes — analysis of deepfake technology in the 2024 election, forensic detection techniques, and real-world examples
- Deepfakes, Detection, and the Misinformation Landscape — comprehensive talk on generative AI, deepfakes, detection techniques, and the weaponization of synthetic media
Key contributions¶
Digital forensics and media manipulation detection: Pioneering techniques for identifying manipulated media by analyzing deviations from physical reality (perspective geometry, shadow consistency, temporal mannerisms).
Behavioral biometrics: Detection methods based on individual speaking patterns, head movements, and facial action units—creating personal "signatures" that deepfakes struggle to replicate over long temporal windows.
Media authentication standards: Advocacy and work around the C2PA (Coalition for Content Provenance and Authenticity) standard for cryptographic authentication of media at the point of capture.
Threat assessment and policy: Extensive analysis of weaponization use cases—from non-consensual intimate imagery to financial fraud, market manipulation, geopolitical deception, and foreign interference in hiring and security.
Topics¶
- Deepfakes, Deepfake Detection, Synthetic media, Synthetic Media Detection, Media Forensics, Digital media forensics, Multimedia Forensics, Media Authentication, Misinformation, Disinformation, Computational Propaganda
Notes¶
Farid's research is grounded in the insight that generative AI, while statistically sophisticated, lacks knowledge of 3D physics and geometry. This asymmetry between human visual perception and physical constraints is the foundation of many detection techniques. His work spans both reactive (forensic) and proactive (authentication standards) approaches, and he emphasizes the need for regulation and responsibility at multiple levels: individual media consumption habits, platform design, AI company safety practices, and government policy.