Good addition, but there's little chance this will work out in practice.

Humans with morals follow rules, sometimes. Probabilistic software acting autonomously or following commands from amoral humans doesn't.