Underground CO2 storage is crucial for sustainability as it reduces greenhouse gas (GHG) emissions, helping mitigate climate change and protect the environment. This research explores the use of Explainable Artificial Intelligence (XAI) to enhance the predictive modelling of CO2 solubility in brine solutions. Employing Random Forest (RF) models, the study integrates Shapley Additive exPlanations (SHAP) analysis to uncover the complex relationships between key variables, including pressure (P), temperature (T), salinity, and ionic composition. Our findings indicate that while P and T are primary factors, the contributions of salinity and specific ions, notably chloride ions (Cl−), are essential for accurate predictions. The RF model exhibited high accuracy, precision, and stability, effectively predicting CO2 solubility even for brines not included during the model training as evidenced by R2 values greater than 0.96 for the validation and testing samples. Additionally, the stability assessment showed that the Root Mean Squared Error (RMSE) spans between 8.4 and 9.0 for 100 different randomness, which shows good stability. SHAP analysis provided valuable insights into feature contributions and interactions, revealing complex dependencies, particularly between P and ionic strength. These insights offer practical guidelines for optimising CO2 storage and mitigating associated risks. By improving the accuracy and transparency of CO2 solubility predictions, this research supports more effective and sustainable CO2 storage strategies, contributing to the overall goal of reducing greenhouse gas emissions and combating climate change.