Should AI Be Regulated in Healthcare?

The question of whether artificial intelligence should be regulated in healthcare is no longer theoretical. With thousands of AI-enabled medical devices already cleared by the FDA and the European Union’s AI Act classifying most healthcare AI as “high-risk,” the regulation train has left the station. The real debate now centers on how to regulate AI in healthcare without crushing the innovation that could save millions of lives.

The Case for Strong Regulation

Healthcare AI is not a consumer app where a bad recommendation means a wasted evening watching a mediocre film. A flawed algorithm can misdiagnose cancer, recommend the wrong drug dosage, or systematically underserve an entire demographic. The FDA’s Software as a Medical Device (SaMD) framework attempts to address this by categorizing AI tools based on the seriousness of the condition they target and the significance of the information they provide to clinical decisions. Under this framework, an AI tool that flags potential sepsis in the ICU faces far more scrutiny than one that helps schedule appointments — and rightly so.

The Innovation Dilemma

Yet the regulatory landscape creates genuine challenges for innovation. The traditional medical device approval process assumes a product is “locked” at the time of clearance. AI, by its nature, is designed to learn and improve over time. A continuously learning algorithm that gets better with each patient interaction is fundamentally different from a static piece of software — but current regulations often require a new submission for every significant update. This creates a perverse incentive: companies either freeze their models (sacrificing the core advantage of AI) or face an endless cycle of regulatory submissions. The FDA’s Predetermined Change Control Plan is an attempt to address this, allowing manufacturers to describe anticipated modifications in advance, but it remains a work in progress.

International Harmonization

The challenge is compounded by the patchwork of international regulations. The EU AI Act, the FDA’s evolving framework, Health Canada’s guidance, and the UK’s MHRA approach all differ in meaningful ways. For a company developing an AI diagnostic tool, navigating this regulatory maze can cost millions and add years to deployment. International harmonization efforts through forums like the International Medical Device Regulators Forum (IMDRF) are making progress, but a truly unified global framework remains distant. In the meantime, patients in less-regulated markets may be exposed to tools that would never pass muster in the US or EU, while patients in heavily regulated markets wait years for tools that could help them today.

Finding the Balance

The path forward likely involves a risk-proportionate approach that combines pre-market scrutiny with robust post-market surveillance. Low-risk administrative AI tools could operate under streamlined notification pathways, while high-risk clinical decision support systems would require rigorous pre-market validation including evidence of performance across diverse populations. All healthcare AI, regardless of risk tier, should be subject to ongoing real-world monitoring that can detect performance degradation, emerging biases, or safety signals. This is not a theoretical framework — it is essentially what the most thoughtful regulators are already building toward.

The Stakes

Getting regulation right is not just a policy exercise. It is a patient safety imperative. Too little regulation and we risk repeating the mistakes of past medical technologies that were adopted enthusiastically before their harms became apparent. Too much regulation and we condemn patients to worse outcomes because beneficial tools never reach them. The answer is not to choose between innovation and safety but to build regulatory systems sophisticated enough to deliver both. That requires regulators who understand AI, developers who understand clinical workflows, and clinicians who are willing to engage with the messy details of both.