The field of remote sensing is witnessing significant advancements in out-of-distribution (OOD) detection and vision-language modeling. Researchers are developing innovative methods to improve the robustness and accuracy of remote sensing models, particularly in detecting novel or anomalous patterns. A key direction is the integration of vision-language modeling, which enables the alignment of visual and textual features to enhance OOD detection and semantic segmentation. Notably, the use of multimodal large language models (MLLMs) is becoming increasingly popular, as they can provide expressive negative sentences to characterize OOD distributions and improve detection performance. Another important trend is the development of efficient and training-free methods, which can reduce the computational cost and improve the scalability of OOD detection models. Noteworthy papers include: DGL-RSIS, which proposes a training-free framework for remote sensing image segmentation by decoupling global spatial context and local class semantics. RS-OOD, which introduces a vision-language augmented framework for OOD detection in remote sensing, leveraging spatial feature enhancement and dual-prompt alignment. ANTS, which shapes an adaptive negative textual space using MLLMs to enhance OOD detection. Efficient Odd-One-Out Anomaly Detection, which proposes a DINO-based model that reduces parameters and training time while maintaining competitive performance. Polysemantic Dropout, which proposes a novel inference-time OOD detection algorithm for specialized LLMs using Inductive Conformal Anomaly Detection. Feed Two Birds with One Scone, which proposes a novel regularization to constrain the distance of fine-tuning and pre-trained models in the function space, aiming to preserve OOD robustness.