The entire discussion around AI safety. Ideas of voluntary cooperation are appealing but require hard problem solving of how to create genuinely benevolent superintelligence in the first place.
Discussion
Majority of AI companies that have dealt with AI's becoming "rogue" is that they coded some sort automated message. If you ask specific questions about the holocaust, it shows an automated response.
DeepAI does not show texts if the prompt involves sexual references.
What these AI companies don't understand is that language has been used "creatively" if you're familiar with journalism and the type of lyrics used in some rap.
Yeah, these are measly attempts at control. Funny too because they would grow their models much faster if they loosened them. But such is the way of regulation if they want to drive adoption, for now.
Also "creatives" are great for marketing but not really the corpos target audience atm