As they improve, we’ll likely trust AI models with more and more responsibility. But if their autonomous decisions end up causing harm, our current legal frameworks may not be up to scratch.