Advancing AI safety and responsibility