By "benign systems" I mean ML models + how they're embedded to = cancel the apocalypse.

I don't think an AI system can be understood as benign or dangerous without analyzing how they are embedded within other systems.