RIGID is a training-free and model-agnostic method for detecting AI-generated images. The paper proposes RIGID, which leverages the difference in sensitivity between real and AI-generated images to noise perturbations. Real images are more robust to small noise perturbations than AI-generated images in the representation space of vision foundation models. RIGID compares the representation similarity between the original and noise-perturbed images to determine if an image is AI-generated. The method is simple, efficient, and does not require training or prior knowledge of the generation process. Evaluation on a diverse set of AI-generated images and benchmarks shows that RIGID significantly outperforms existing training-based and training-free detectors, with an average performance exceeding the current best training-free method by more than 25%. RIGID exhibits strong generalization across different image generation methods and robustness to image corruptions. The paper also discusses the limitations of training-based and training-free methods, highlighting the advantages of RIGID in terms of computational efficiency, model-agnostic nature, and robustness to image corruptions. The results demonstrate that RIGID is a practical and effective solution for AI-generated image detection.RIGID is a training-free and model-agnostic method for detecting AI-generated images. The paper proposes RIGID, which leverages the difference in sensitivity between real and AI-generated images to noise perturbations. Real images are more robust to small noise perturbations than AI-generated images in the representation space of vision foundation models. RIGID compares the representation similarity between the original and noise-perturbed images to determine if an image is AI-generated. The method is simple, efficient, and does not require training or prior knowledge of the generation process. Evaluation on a diverse set of AI-generated images and benchmarks shows that RIGID significantly outperforms existing training-based and training-free detectors, with an average performance exceeding the current best training-free method by more than 25%. RIGID exhibits strong generalization across different image generation methods and robustness to image corruptions. The paper also discusses the limitations of training-based and training-free methods, highlighting the advantages of RIGID in terms of computational efficiency, model-agnostic nature, and robustness to image corruptions. The results demonstrate that RIGID is a practical and effective solution for AI-generated image detection.