New AI Method AIFIND Improves Detection of Evolving Face Forgery Techniques

A research paper introduces AIFIND, a novel approach for Incremental Face Forgery Detection that addresses catastrophic forgetting in AI systems. The method uses semantic anchors derived from low-level artifact cues to stabilize learning as new forgery types emerge. Unlike previous approaches that rely on data replay or binary supervision, AIFIND explicitly constrains the feature space through Artifact-Probe Attention, which aligns volatile visual features with stable semantic anchors. The system includes an Artifact-Driven Semantic Prior Generator to create invariant semantic anchors, establishing a fixed coordinate system. An Adaptive Decision Harmonizer preserves angular relationships between classifiers. This technical advancement responds to the continuous emergence of new forgery types, making incremental detection a crucial paradigm. The paper was published on arXiv with identifier 2604.16207v1 under the cross announcement type.