increase global transparency at the frontier, asking questions like: Does the system show signs of being able to self-improve capabilities? Can it specify its own goals? Can it acquire more resources without human oversight? Is it deliberately trained in deception or manipulation?
