Abstract: The Healthy Technology Act of 2025, a bill being considered by Congress, would permit artificial intelligence (AI) systems to prescribe medications without human sign-off. While allowing AI to practice independently may still be a stretch, large language models are already being used to support diagnosis and treatment.1 These applications challenge the current regulatory framework for clinical AI, which has focused on software as a medical device (SaMD). The US Food and Drug Administration (FDA) has cleared more than 1000 AI tools using this framework, mostly for narrow, well-defined tasks. However, generative models can be applied across domains, straining the SaMD framework. In this Viewpoint, we explore the application of a licensure paradigm to clinical AI systems, arguing that this offers a more reliable regulatory framework.