In brain tumor surgeries, image-guided neurosurgical systems (or neuro-navigation systems, respectively) help neurosurgeons to locate tumors by aligning preoperative image data with the patient's coordinate system. However, brain deformation, aka brain shift, can lead to misalignment between estimated and actual tumor positions and increase the risk of brain damage. Existing methods for the compensation of brain shift often involve additional intra-operative imaging (MRI or ultrasound) or are initiated by manual landmark registration, both of which disrupt the surgical workflow. In the digibrAIn project, we propose a novel computational approach that uses intraoperative image data from the stereomicroscope after opening the skull (craniotomy) and the dura mater to perform automatic, AI-based non-rigid registration without interference with the established clinical workflow. Our framework utilizes advanced deep learning techniques for non-rigid 3D point cloud registration, combined with synthetic training data generated by Finite Element Method (FEM) to realistically represent potential brain structure deformations. Additionally, it takes advantage of transfer learning to enable the time-consuming initial FEM simulations and AI model training based on the diagnostics MRI weeks before surgery, and then upgrade the model based on the most recent information from a planning MRI the day before surgery in a fast and efficient way. As part of the project, the proposed approach will be verified using a novel instrumented brain phantom by comparing measured and predicted brain shifts. This framework has the potential to accurately predict real-time brain shift, support the planning of optimal surgery trajectories and guide surgeons by providing updated information of tumor location and critical brain structures without disrupting the workflow. Ultimately, our approach aims to enhance surgical quality, improve patient safety, and positively impact patient’s quality of life.