Cloud computing allows a computationally weak user to delegate computation to a powerful server. The unprecedented accessibility of enormous computing power is not always used for good: malicious users may take advantage of powerful cloud servers to carry out ill deeds, as demonstrated by the 2011 Sony hack that was launched from the Amazon EC2 cloud platform. This highlights the importance of responsible delegation of computation services that actively attempt to prevent malicious use of their platform.
Meanwhile, recent cryptographic developments are enabling mph{private} delegation of computation, allowing users to delegate computation on sensitive data without disclosing anything about it to the server. Responsibility and privacy seem to be at odds: how can we expect cloud services to prevent misuse of their platform if they cannot even look at the delegated computation?
In this work, we devise a theoretical framework to model misuse-prevention by cloud services, offering a rigorous definition of mph{responsible} delegation of computation. We then give an answer to the above question by presenting a cryptographic compiler that takes any (publicly-verifiable) one-round delegation protocol and produces a responsible and private one-round delegation protocol, using fully-homomorphic encryption and non-interactive zero-knowledge arguments.