Deepfake detectors' accuracy declines with untrained videos

A new study reveals that deepfake detectors are struggling to identify fake videos in real-world scenarios, achieving accuracy rates of only 66% on untrained data. This highlights significant vulnerabilities in current detection technology. Researchers found that while detectors perform better on controlled datasets, their effectiveness drops significantly when faced with newer deepfakes from the internet. The best-performing detector identified fakes correctly 69% of the time, down from 86% on older datasets. Experts suggest that specialized detectors may improve accuracy, but they still face challenges in adapting to rapidly evolving deepfake technology. Increased public awareness and better regulatory measures are also needed to address the growing issue of deepfakes.