Terry Garton I’m very happy. The Bipartisan Policy Center has released an interesting new report. What I’m going to talk about now is the regulation of medical AI tools. When I think of health AI tools, when I think of what the average person thinks of, I think of fitness trackers and wearables. What are we really talking about here? What’s in this report?
maya sandrow Yes, absolutely. So my organization, the Bipartisan Policy Center, has done a lot of work to answer that very question. AI is used in a variety of ways throughout healthcare delivery. To keep things simple, I’ll categorize them into two broad buckets. And we’ll call the first bucket Management AI. It is a type of AI used for operational and administrative purposes. So, as a patient, you might interact with this to set your schedule or navigate your benefits. One of the fastest growing administrative AI applications today is used by healthcare providers to record patient conversations. They are called AI or ambient scribes. And health insurance companies also use something called pre-authorization for insurance claims. So these are the entire operational uses of AI. And then there’s the clinical bucket. There are also conventional medical devices. It is AI that serves as a tool used to diagnose, prevent, and treat diseases. And then there are all the other kinds of tools that people are wearing these days, like rings and Apple watches. As a result, there are a wide variety of tools available. Even among medical devices, there are various applications. The most common one so far is through medical imaging, so we’ll explain all of them in detail for your convenience.
Terry Garton That’s a nice setup, but what sticks in my mind there is why ask about regulation? What’s wrong with regulating all the different health-related AIs?
maya sandrow Well, there are different regulatory questions depending on the type of AI tool, right? So the Food and Drug Administration, which is an agency within the Department of Health and Human Services (HHS) at the federal level, regulates medical devices to make sure they’re safe and effective before they go on the market, right? Therefore, the same questions apply to AI used in clinical functions. We want to make sure they are safe and effective before we use them on patients. And you can imagine, for example, that you might need policy and oversight for AI that’s used to determine whether your medical procedure is covered by your health insurance, right? So, depending on the type of AI application, there are different policy questions.
Terry Garton One of the findings from this report is that, as we just discussed, many of these tools are not classified as medical devices. As a result, it may not be clear who has regulatory responsibility, or more than one agency may have regulatory responsibility. What’s the stadium like here?
maya sandlow Well, it’s a complex competitive environment. That’s for sure. The same AI may be regulated by different entities depending on where it is used and the data used for training. We map out the most common federal agencies that may play a role. Therefore, the Centers for Medicare and Medicaid Services may play a role in coverage and reimbursement decisions related to AI. That is, the extent to which healthcare providers are reimbursed or patients are reimbursed for AI used in clinical care. The Office of Civil Rights within HHS oversees something called HIPAA, which is important for patient privacy. That means any AI used for traditional health care delivery, patient data, or privacy issues would be overseen by that department. That doesn’t necessarily apply to the rings and watches people wear, and it doesn’t necessarily apply to traditional healthcare delivery systems. It’s complicated, but that’s not always the case. The Federal Trade Commission may become involved if there is misleading marketing regarding the intended use of these tools. Simply put, the Office of the National Health IT Coordinator is another body that has certified electronic health records. So, a lot of patients are familiar with these electronic health records, a lot of health care providers are using these electronic health records, and if certain types of AI are built into those tools, they’re empowered to do that. In short, the web is extremely complex, and developers and providers must navigate different types of AI tools. And that’s at the federal level. States are also seriously involved, so the same tool may have strict regulations in one state and no regulations in another.
Terry Garton Maya Sandalow is deputy director of the Health Program at the Bipartisan Policy Center. Maya, I think what you just described is beyond most people’s imagination. And in the healthcare sector, AI appears to be being adopted more rapidly than the regulatory framework to govern it. Everyone is implementing AI in some form, but perhaps not all the rules have been written yet. What gaps and risks does this create?
maya sandrow That’s a great and important question. I think one of the central trends to keep in mind, related to your question, is that the most common form of AI used in healthcare today is for these administrative purposes. This means doctors can use it to transcribe notes, and insurance companies can use it to review medical claims to understand what will be reimbursed and covered for patient providers. There are many possibilities for clinical applications of AI. AI has a lot of potential to help prevent diseases earlier, diagnose more accurately, and treat the most complex diseases. And at the moment, the regulatory framework is very complex and the payment mechanisms are not very clear. In other words, it has a kind of chilling effect. So I think there’s a big policy question of how can we create regulatory frameworks and payment policies to ensure that these safe, effective and high-value solutions actually reach the patients who need them most?
Terry Garton Who will take the lead in developing that policy framework?
maya sandlow I think a lot of states are trying to think about this right now, right? And because states are laboratories of democracy, there may be some best practices and lessons that can be learned. But at the same time, these technologies operate at a national level, and the same developers may be looking to deploy tools across the country. So a lot of people think we need a federal framework, right? And HHS is a federal agency that oversees health care delivery in a number of ways. And they’re thinking about it very actively now.
Terry Garton In the meantime, opportunities for industry to benefit tend to be in these regulatory gaps, right? So how are incentives for industry affecting the regulatory process?
maya sandrow Well, I think this is very important. The goal is to create a regulatory framework that developers get right and that technology companies are encouraged to go through rather than around. At the moment, it’s very complicated to get through that, and there’s a lot of incentive to say that what I’m manufacturing, what I’ve developed, is not a medical device because there’s no clear payment while getting through it. It should not be regulated by the FDA. There are various discussions, right? For example, users of AI tools such as chat bots may not be aware that they are unregulated and not intended to be used for diagnostic or therapeutic purposes. So, to ensure that patients have safe and effective tools, we need to create frameworks designed to help developers do just that.
Terry Garton This safe and effective part comes in because another function of the regulatory environment is to hold people accountable when things go wrong.
maya sandrow absolutely. That’s a big issue now, responsibility, right? In fact, when AI is used in healthcare today, healthcare providers are often held accountable if something goes wrong. Countries are actively considering this. There are important policy developments to watch in this area. The ONC Office of the National Health IT Coordinator proposed this model card or nutrition labeling policy several years ago. Originally, developers needed to provide information about how the AI worked compared to what people saw on nutrition labels. And that allows health care providers to be aware of that and share more of the risk. In early January, the new government proposed repealing it. While there are supporters and opponents of this proposed rule, I would like to point out that some health care providers are very concerned that this rule could remove the tools they need to make decisions about how AI is used in health care and could increase liability concerns.
Terry Garton While we’re in this kind of regulatory logjam, what do individuals need to know to not only protect their privacy and data, but also to take advantage of the power of these AI tools that are out there?
maya sandlow That’s a great question. I think the most basic answer is to not assume that the AI solutions you are interacting with are regulated in the same comprehensive way as the drugs you may be taking. That’s not to say there aren’t powerful ways to manipulate these solutions, but keep in mind that they are not healthcare providers. And honestly, I think it’s also a good idea to talk to your healthcare provider about the tools you’re using. And then plan together on how to actually use these in ways that enhance care.
Terry Garton And where do you think federal regulation will go next?
maya sandlow Definitely something to keep an eye on. I think there is bipartisan agreement on the need for a federal framework because of the patchwork of states. The devil is in the details, and that’s where it’s hard to reach bipartisan agreement, right? But the need for a federal framework will continue to be debated for years to come, as will ensuring that executive-level agencies have the powers and resources they need to play a greater role within that framework.
Copyright © 2026 Federal News Network. Unauthorized reproduction is prohibited. This website is not directed to users within the European Economic Area.
