[R] what are some important research areas for AI safety?

I have been looking into it and have been asking myself, in 2026 what would be/are the most critical research questions that are understudied or should be answered urgently?