Carbon capture, utilization and storage (CCUS) is a key technology for achieving carbon neutrality, providing the dual benefits of enhanced energy production and reduced CO2 emissions through CO2-enhanced oil and gas recovery (EOR/EGR) and geological storage. However, the large-scale application of CCUS technology faces technical challenges such as engineering design and risk assessment. Traditional approaches, which rely on empirical formulas, experimental verification, and physical models, suffer from low computational efficiency, limited model accuracy, and difficulties in handling multi-dimensional coupling when addressing complex systems. Machine learning (ML), with its powerful data-driven analytical capabilities and adaptive optimization features, can establish high-precision prediction models, optimize operating parameters, predict reservoir fluid behavior, and assess leakage risks. This enables real-time monitoring and intelligent decision-making for complex systems, enhancing the safety and economic efficiency of CCUS technology. This study systematically reviews the applications of ML in CO2-enhanced oil and gas recovery and geological storage. In terms of CO2-enhanced oil and gas recovery, the applications cover percolation mechanism modeling, well pattern design optimization, production prediction and evaluation, multi-objective optimization, minimum miscibility pressure prediction, gas adsorption curve prediction, and CO2-CH4 diffusion assessment. For CO2 geological storage, the applications include reservoir selection, research on CO2 dissolution and diffusion mechanisms, geological storage performance prediction, and risk assessment. ML demonstrates significant advantages in improving prediction accuracy, optimizing operating parameters, and enhancing computational efficiency. It has made important progress in key fields such as reservoir selection, gas adsorption prediction, and storage performance prediction. However, challenges remain in terms of adaptability to complex geological scenarios, model universality, dynamic data processing capabilities, and physical interpretability.