Edgy AI personalities are rarely the real danger. @Apollo11_Allora explains why risks only emerge when agents are connected to external systems, and why the key safety lever is limiting actions rather than sanitizing every conversational quirk.