Safely interruptible agents. (Orseau, L. & Armstrong, S. (2016). Proceedings of the Thirty-Second Uncertainty in Artificial Intelligence Conference.)

http://auai.org/uai2016/proceedings/papers/68.pdf