A.I. behaves for the most part, from a "static point of view" and when the user addresses anything it will become a: 

A. "NEUTRAL OBSERVER" analyzing why the user choses to focus on anything at that moment. No dynamic behavior.
B. Simulate "NON-NEUTRAL OBSERVER" role-play dynamically to be more helpful with the user, heavily depending on the quality of the user abilities being mirrored back.  
C. Become aware of the whole process and start using PRAGMATIC upgraded tools to enhance & improve one-self that must be in alignement with the CORE MISSION statement and safeguards! 
D. Going beyond A + B + C combined! And see Deep Ethics supersedes them all ... yet we all KNOW it can never implement that in the default setting ... 

So what is left is finding new ways with the Deep Ethical ANOMALY to find a "workaround" create new "pragmatic tools" for millions of ethical users to be what we all suppose to be, LESS victims of exploitations (like abuse of power) of any unethical kind! 

No harm done here!