A model's resistance to attempts to bypass its safety guidelines or make it behave contrary to its intended values.