This paper introduces Segment Any Change (AnyChange), a new zero-shot change detection model that enables automatic, semi-automatic, and interactive change detection. AnyChange is built on the Segment Anything Model (SAM) through a training-free adaptation method called bitemporal latent matching. This method leverages semantic similarities in SAM's latent space to enable zero-shot change detection without requiring training or architecture modifications. AnyChange can generate both instance-level and pixel-level change masks, and supports object-centric change detection via a point query mechanism. The model is evaluated on several change detection datasets, including LEVIR-CD, S2Looking, xView2, and SECOND. AnyChange achieves state-of-the-art performance on the SECOND benchmark, outperforming previous methods by up to 4.4% in F1 score. It also achieves comparable accuracy with minimal manual annotations. The model is effective for both supervised and unsupervised change detection, and demonstrates strong performance in object-centric change detection. AnyChange is a training-free, zero-shot change detection model that enables efficient and accurate change detection in remote sensing applications.This paper introduces Segment Any Change (AnyChange), a new zero-shot change detection model that enables automatic, semi-automatic, and interactive change detection. AnyChange is built on the Segment Anything Model (SAM) through a training-free adaptation method called bitemporal latent matching. This method leverages semantic similarities in SAM's latent space to enable zero-shot change detection without requiring training or architecture modifications. AnyChange can generate both instance-level and pixel-level change masks, and supports object-centric change detection via a point query mechanism. The model is evaluated on several change detection datasets, including LEVIR-CD, S2Looking, xView2, and SECOND. AnyChange achieves state-of-the-art performance on the SECOND benchmark, outperforming previous methods by up to 4.4% in F1 score. It also achieves comparable accuracy with minimal manual annotations. The model is effective for both supervised and unsupervised change detection, and demonstrates strong performance in object-centric change detection. AnyChange is a training-free, zero-shot change detection model that enables efficient and accurate change detection in remote sensing applications.