How Researchers Measure, Detect and Benchmark AI Manipulation

Researchers from the University of Kent deliver one of the most comprehensive deepfake reviews to date, covering English and Chinese literature. The paper clarifies competing definitions, breaks down detection metrics like AUC, EER and F1-score, maps major datasets such as Deepfake Detection Challenge, and analyzes global standards efforts from ISO/IEC. It also meta-reviews 12 key surveys, highlighting open challenges in benchmarking, realism, and regulation.