Friendly Fire: UAlbany Neural Network Learns to Catch AI Fakes

Generative networks have pushed the problem of disinformation to the limit, and on March 8, 2026, researchers at the University at Albany presented a "cure." Their new system is designed to automatically detect deepfake content, including generated military images and videos.

The technology is likely based on the principle of Generative Adversarial Networks (GANs), where a detector is pitted against a generator. A key feature of the UAlbany system is its high efficiency (few-shot learning) on small datasets. This means the algorithm doesn't need millions of pictures to understand that the pixel textures of a generated soldier in the background have "blurred." The battle of algorithms is moving to a new level: AI detectors are becoming a mandatory filter in social networks and news agencies, implementing the concept of "Zero Trust" regarding any media content.

Source: UAlbany / Times Union