Magnetic resonance imaging (MRI) is an imaging modality able to offer invaluable insights into human pathology and anatomy, thus being an important diagnostic tool in healthcare. AI-powered tools are already available to assist in image acquisition, reduce scan times, automate slice prescription and patient positioning, optimise protocol selection, reduce image noise, and streamline workflows. MR image reconstruction techniques based on deep learning (DL) have already shown superior performance compared to conventional reconstruction methods, allowing for enhanced image quality, reduced acquisition times, and improved diagnostic accuracy. However, adoption of AI tools in clinical MR practice requires careful consideration of ethical and legal challenges associated with the use of AI; data privacy, transparency, and explainability of AI methods should be ensured, and novel techniques, such as federated learning, must be employed to mitigate ethical risks and safeguard data sharing. Multimodal AI will be a key player in the future use of AI in MRI, since this will allow better patient outcomes and personalised patient care for better patient experience.