This allocation will support two distinct but complementary research projects conducted under my supervision at Halmstad University. The first project focuses on Multilingual Healthcare Benchmarking for Swedish, including the development and validation of Swedish medical benchmarks, multilingual evaluation of Large Language Models, model adaptation and distillation, and longitudinal ICD prediction. The second project focuses on Robustness in Vision-Language Models and Safety Alignment, including adversarial robustness of multimodal models, safety alignment, task gaming, analysis of behavioral and execution traces, and evaluation of AI models and agents under adversarial and complex task settings. Both research activities are being carried out within my research group and require substantial GPU resources for model training, large-scale inference, repeated evaluation, adversarial experimentation, and model adaptation. I am therefore consolidating the computational requirements of these two research directions into a single NAISS allocation, so that the requested resources can be shared between the projects according to their experimental needs. This consolidation follows the recommendation that the computational requirements associated with projects under the same PI should be collected within one proposal. Possible continuation: The allocation may also support closely related follow-up studies that emerge from these two research directions during the project period.