Veritas: Generalizable Deepfake Detection via Pattern-Aware Reasoning

Researchers introduce Veritas, a multi-modal large language model designed for deepfake detection that uses pattern-aware reasoning to mimic human forensic processes. The system addresses real-world challenges through the HydraFake dataset and achieves significant improvements in detecting unseen forgeries across different domains.