New Study Finds AI System Outperforms Human Examiners in Video Authentication Tests

A new forensic study has found that an artificial intelligence system performed better than human examiners in a controlled video authentication test, raising questions about how AI could support the examination of digital evidence.

The research, announced by Magnet Forensics on October 26, 2026, compared an automated expert system with trained forensic examiners tasked with determining whether video files were authentic or manipulated.

Published in the Journal of Forensic Sciences, the study evaluated Magnet Verify, a forensic tool designed to help investigators interpret media authentication results.

Researchers assembled 14 video files representing different scenarios, including original camera recordings, edited videos, social media content, AI-generated media, and deliberately challenging examples.

Twenty trained examiners assessed the files independently. Their conclusions were then compared with those generated by the automated system and checked against the known origins of the videos.

According to the company, the AI expert system achieved 91% accuracy, compared with an average of 69% among human examiners.

The automated system also reportedly made no incorrect classifications, declining to provide definitive conclusions when the available evidence was insufficient.

These findings highlight a growing challenge in digital forensics. As manipulated and AI-generated videos become increasingly sophisticated, determining whether digital media has been altered requires careful technical examination.

Forensic investigators often examine metadata, compression patterns, encoding characteristics, and other digital artifacts to evaluate a recording's authenticity.

However, the study's relatively small dataset limits how broadly its findings can be applied. Performance across larger collections, different manipulation techniques, and unfamiliar recording environments remains uncertain.

The results also come from research associated with the company developing the technology, making independent replication particularly important.

Rather than establishing that AI can replace forensic examiners, the study suggests that automated interpretation could become a useful supporting tool when its accuracy, limitations, and underlying methods are properly validated.

For courts and investigators, the central question remains whether AI-assisted conclusions can be independently examined, reproduced, and supported by reliable forensic evidence.