A framework for thinking about AI power-seeking — AI Alignment Forum