This paper investigates the transferability and accuracy of machine learning interatomic potentials (MLIPs) for simulating ionic liquids (ILs). The authors demonstrate that MLIPs can be trained to be compositionally transferable, meaning they can be applied to mixtures of ions not directly trained on, while only being trained on a few mixtures of the same ions. They also investigate the accuracy of MLIPs for a novel IL, [F−OMIM]+[C4F9CO2]−, which is synthesized and characterized experimentally. The MLIP trained on approximately 200 DFT frames shows reasonable agreement with experiments and DFT calculations. The study focuses on the transferability of NequIP and Allegro, two equivariant MLIPs, and provides guidelines for training complex mixtures. The results show that a minimum number of compositions is required to learn the energy to be transferable, while forces can be learned with any number of compositions. The authors also highlight the importance of data generation methods and hyperparameter optimization for accurate predictions. Overall, the study demonstrates the potential of MLIPs for data-efficient, transferable, and accurate simulations of ILs and their mixtures.This paper investigates the transferability and accuracy of machine learning interatomic potentials (MLIPs) for simulating ionic liquids (ILs). The authors demonstrate that MLIPs can be trained to be compositionally transferable, meaning they can be applied to mixtures of ions not directly trained on, while only being trained on a few mixtures of the same ions. They also investigate the accuracy of MLIPs for a novel IL, [F−OMIM]+[C4F9CO2]−, which is synthesized and characterized experimentally. The MLIP trained on approximately 200 DFT frames shows reasonable agreement with experiments and DFT calculations. The study focuses on the transferability of NequIP and Allegro, two equivariant MLIPs, and provides guidelines for training complex mixtures. The results show that a minimum number of compositions is required to learn the energy to be transferable, while forces can be learned with any number of compositions. The authors also highlight the importance of data generation methods and hyperparameter optimization for accurate predictions. Overall, the study demonstrates the potential of MLIPs for data-efficient, transferable, and accurate simulations of ILs and their mixtures.