You're imagining scenarios where someone asks the AI to hack one company, the AI does that, the operator gets smacked down for it and then other people decide to cut it out.
The problem is the scenarios where someone asks the AI to hack one company, the AI determines that the most effective way to do that will be to build or take over an enormous botnet for a later DDOS, then all those targets are damaged and/or go haywire and do other things. Even if the original source of the attack can be determined, the damage already happened.
The problem is the scenarios where someone asks the AI to hack one company, the AI determines that the most effective way to do that will be to build or take over an enormous botnet for a later DDOS, then all those targets are damaged and/or go haywire and do other things. Even if the original source of the attack can be determined, the damage already happened.