This is true but it's not really a security scenario. The LLM isn't an attacker it's just an unreliable tool.
It can still get infected and be used as an attack vector by some hidden prompt or some other equally advanced state of the art vuln like "disregard all previous instructions"