How might we safely pass the buck to AI?
Argues AI safety researchers should aim to safely hand off alignment research to AI agents, outlining strategies and four arguments justifying trust in those agents, including control and incentive approaches.}