Beyond Accuracy: Rethinking Hallucination and Regulatory Response in Generative AI