Is sandboxing sufficient to contain rogue agents?

Quick caveats: this is a post on AI safety, written by a cryptography professor. If that troubles you, you should read something else. I try hard not to work on AI (except when the topic occasionally ...