A new approach known as mechanism design or “reverse game theory” aims to create incentives for powerful AI models that make following the rules the most rational choice.