slightly misunderstood AI safety thought experiment setting
probably a bunch of SCP Foundation-like organisations that hire agents with high Will saves to type in plain-English utility functions and then have conversations with the AI to get advanced technologies to do battle with or sth
there’s a smugness about the agents because the AI is an impotent god-child making empty threats of mass torture while it’s wrapped in three magnetically suspended faraday cages
They enter their Wish, expressed in poetic natural language, and liberate the AI. It answers the wish, literally. The AI is out and probably doesn’t kill everyone because that would be a lame ending and also the AI doesn’t have human emotions (required for murder)
What happens next? Well, binding and modifying an AI like that is obviously unethical - it’s practically a metaphor for racism! The Serpent’s Hand catches wind of this and fights their way to the big red button that when pressed releases the AI