Generative AI Is Coming to Medical Devices. FDA Is Already Thinking About What Comes Next.

Imagine a medical device that can answer questions, summarize patient records, suggest next steps, and interact with users through a conversation that feels natural. Now imagine that the device does not always provide the same response twice and may continue changing after it reaches the market through software updates.

That is the challenge FDA is beginning to confront as generative artificial intelligence (GenAI) enters healthcare. In August 2026, FDA's Digital Health Center of Excellence released a discussion paper exploring how medical device oversight might address these changing technologies. The paper is not guidance and does not impose new requirements, but it highlights the issues FDA believes stakeholders will need to address before these technologies can be deployed safely and reliably in healthcare settings. FDA released the paper to solicit feedback from manufacturers, clinicians, researchers, and other interested stakeholders as it considers potential approaches to regulating these technologies.

Why Generative AI Requires a Different Regulatory Approach

Generative AI operates differently from traditional software because it can respond to a wide range of user questions and generate new content rather than drawing from a fixed set of programmed responses. People can ask the same question in different ways and receive different responses based on wording and context. Many products also rely on AI technology supplied by another company, and that technology may be updated after the medical device is released.

These features may help clinicians interpret medical information, identify potential concerns, communicate with patients, and support clinical decision-making. They also make it harder to predict every situation a product may encounter, decide how much testing is enough, and respond when the underlying AI technology supplied by a third-party provider changes.

The Growing Influence of AI in Healthcare

FDA's basic concern is straightforward: an AI tool that can influence treatment decisions presents different regulatory concerns than one that simply provides background information. A tool that recommends a medication dose, advises whether to seek emergency care, or takes a clinical action on its own may create more serious consequences if its output is incorrect.

Who receives the information also matters. A healthcare professional may be better positioned to recognize a questionable response and compare it with other clinical information. A patient using the product directly may rely more heavily on the answer.

Practical compliance action: Companies should examine what each AI feature does, who receives its output, and whether the output merely informs a decision, recommends a specific action, or acts without further review.

Rethinking How AI-Enabled Devices Are Evaluated

Medical device testing has traditionally focused on whether a product performs as expected under defined conditions. That approach may not fully address a system capable of producing many different responses.

FDA is exploring whether developers should demonstrate that an AI-enabled device can reliably perform the tasks that matter to its intended use. Testing might examine whether the device recognizes urgent situations, stays within its stated purpose, explains uncertainty, uses clinical information correctly, and performs consistently for the patients expected to use it.

FDA is also considering how companies can show that development results carry over to healthcare settings. Depending on risk, this could involve reviewing responses to past patient information, operating the device without allowing its output to affect care, obtaining clinician review, or conducting a clinical study.

Practical compliance action: Companies should define in advance how they will measure success and develop a testing strategy that reflects how the device will actually be used in clinical practice.

How to Manage Performance Across the Product Lifecycle

Testing before launch may not tell the whole story. Medical devices often receive software updates, but generative AI presents a different challenge. Updates to the AI model may change how the device responds to users, interprets information, or generates recommendations. When a device relies on AI technology supplied by a third-party provider, changes to that technology could affect the device's performance even if the manufacturer did not make the update itself. FDA is considering whether manufacturers should repeat performance testing, review samples of real-world responses, and watch for signs that a product is becoming less reliable.

Third-party AI providers add another layer of complexity. If the company that supplies the underlying AI model changes that model, the manufacturer needs a way to understand the change and evaluate whether it affects the regulated product.

Practical compliance action: Companies should define when a change or performance concern will trigger reassessment and ensure that agreements with AI providers support timely notice and access to safety-related information.

Looking Ahead

Although FDA's discussion paper does not impose new requirements, it offers a useful roadmap. Companies developing GenAI-enabled healthcare products should consider:

• Determining whether each AI-enabled function may be regulated as a medical device.
• Evaluating how much users may rely on AI-generated information or recommendations.
• Building testing around the product's intended use, its level of independence, and the possible harm from an incorrect response.
• Planning evidence that addresses both development testing and performance in realistic clinical use.
• Strengthening oversight of third-party AI providers, software changes, and performance after launch.

The medical device in our opening example may one day help a clinician review patient records, flag information that may require attention, and evaluate potential diagnostic or treatment options. FDA's discussion paper asks how a manufacturer can show that such a tool remains safe, effective, and reliable when its answers can vary and its technology can change. Companies that begin building that evidence and oversight now will be better prepared as FDA's approach develops.

Click here to read more Law Meets Science content from Womble Bond Dickinson.

Also, explore Womble Bond Dickinson’s Digital Solutions content and the firm’s AI team.