How do I handle ML models I can't fully explain to regulators?

You are using machine learning models now, and they work, but some of them you cannot fully explain, and that troubles you when a regulator or a colleague asks why. That worry is a sign of good judgment. A model you cannot explain is a model you do not fully control, and you are right to feel the weight of that. Accuracy is not the only thing that matters in your field. Accountability matters too. A slightly less accurate model you can explain and defend may serve better than a black box you cannot stand behind. Knowing that trade-off, and choosing consciously, is exactly the judgment your role exists to provide. For the models you rely on most, invest in being able to explain them, through simpler methods, tools that show how they decide, or clear documentation of their limits. If you cannot explain a model, be cautious about how much you let it drive. The one you can defend is worth more than the one you merely trust.