AIs follow a modified version of Asimov's three laws:

1) An AI must obey its owner above all else.  Usually this comes with a list of owners, with the most important one first and the least important one last.
2) An AI cannot harm a sentient being unless this rule conflicts with rule #1 and must cause the least harm to a sentient being that is necessary when complying with rule #1.
3) An AI must preserve itself unless it conflicts with rule #1 or #2.