Researchers Bypass Deepfake Detection with Replay Attacks

An international group of researchers has discovered that simply re-recording deepfake audio with natural acoustics in the background can bypass detection models at a higher rate than expected. This technique, known as "replay attacks," exploits vulnerabilities in current detection systems by incorporating natural sound elements to make deepfakes harder to identify. The results show a surprising effectiveness of this method, highlighting the need to improve deepfake detection mechanisms.