Cooperative Inverse Reinforcement Learning (CIRL) — supplies a candidate mechanism for → Shutdown / off-switch problem

The paper does not treat the shutdown problem as needing its own bespoke solution; it treats CIRL as already carrying the needed structure. Immediately after introducing CIRL as a framework where "an agent and a human work together to achieve the human's goals," the text continues: "This concept can be applied to situations where we want to make sure a human is not blocked by an agent from shutting the agent down if it exhibits undesired behavior" (concrete-problems, §"Multi-Agent Approaches:", p. 6). The mechanism is specific: because the agent infers the human's goal rather than optimizing a goal fixed in advance, allowing itself to be shut down is simply what a correctly-functioning CIRL agent does whenever the human's inferred goal favors shutdown, no separate off-switch logic is bolted on. The paper still calls shutdown "an interesting problem in its own right," signaling CIRL as a candidate answer, not a settled one.