May I suggest adding to your list of test cases the blueprints for a non-Friendly AI?

By that I mean any program which is expected to be a General Intelligence but which isn't formally or rigorously proven to be Friendly. (I still haven't come to definite conclusions about the plausibility of an AI intelligence explosion, therefore about the urgency of FAI research and that of banning or discouraging the dissemination of info leading to non-F, but given this blog's history it feels as if this test case should definitely be on the list.)