Constraints usually aren't that hard to implement. For example, "computer worms and viruses that defy extermination" is and always will be a computer security problem, not a philosophical AI problem.