That’s our first line of defense in limiting humans, too.

(With AI, as with humans, we have additional means of control, via imposed restrictions on access to resources and other remedies, should the “bunch of sentences” not produce the desired behavior.)