Researchers have developed a new framework for detecting deepfakes that provides interpretable justifications. The approach, called Feature-robust Augmentation and Evidence-grounded Explanation Optimization, addresses two major limitations of current methods: vulnerability to image quality degradation and factually flawed explanations. To improve robustness, the framework uses diversified augmentation strategies and a mean-teacher architecture to stabilize features against augmentations. For explanation, it guides models to prioritize genuine manipulation traces by learning from chosen-rejected explanation pairs. The approach won first place in an Explainable Deepfake Detection Challenge. --- Why it matters: This matters because current deepfake detection methods often fail to provide meaningful explanations for their predictions, making it difficult for users like forensic analysts to understand the rationale behind the detection. This new framework addresses these limitations and provides a more robust and interpretable approach to detecting deepfakes. Source: https://arxiv.org/abs/2608.20913