Common Q&A from Emergo’s Human Factors webinar on FDA’s 2025 AI Draft Guidance

By Erin Davis, Tess Forton and Julee Henry 

In a webinar presented in February 2026 (available to watch on-demand), Emergo discussed the 2025 draft FDA AI guidance and how it impacts human factors for AI-enabled products, including how AI-specific risks impact use-related risk analysis, what UI design principles the FDA expects, how HF validation may need to be adjusted, and what additional documentation will support an FDA submission. Below are answers to the questions asked most frequently during the webinar.  

What are the foundational differences between this draft FDA AI guidance and the EU AI Act? 

While the EU AI Act doesn’t reference human factors as explicitly as the FDA’s draft guidance, it still includes several expectations that align with key HFE topics, such as: transparency, clear instructions to users about how to use AI outputs, ensuring humans can oversee AI and make correct decisions, and managing the risks long-term as the AI evolves (all themes the FDA guidance emphasizes as well). We do not see any significant differences between the two, although we know that the FDA seems to review the HF elements of submissions in a more nuanced way than some notified bodies, so there may be more scrutiny on the HF work related to evaluating AI-enabled devices submitted in the US. However, we recommend human factors engineers monitor this trend to see how things evolve over time.  

How do you evaluate/test all cognitive/knowledge tasks during development when the system and outputs are changing over time? 

Rather than attempting to test every possible permutation of changing AI outputs, we recommend thinking in terms of categories of outputs. This is similar to HF approaches for testing alarms: while you don’t test every individual alarm condition, you do test representative examples from each category to demonstrate that the presentation, information conveyed and required user response are clear and safe. The same principle applies with AI outputs.  

During development, manufacturers can map out the different types of outputs the system can generate and identify a representative set that captures the variability in both content and clinical decisions. This includes considering not only today’s outputs but also how outputs might change as the model evolves (e.g., whether confidence indicators may be added, whether the explanation format might shift, whether the model will change leading to a different output even if inputs haven't changed). We think that proactively discussing how the model might evolve (e.g., with those developing the model) can help you select the right knowledge tasks for an HF validation. That said, because AI models can change post‑market in ways you might not be able to anticipate, it will also be important to periodically revisit whether new or modified outputs warrant supplemental HF validation. 

Where should the model card go in a software-only device?  

If the device is software only, you could include the model card (see the guidance’s Appendix F) in a section in your app where you present other Help or Support info, or you might also include it in any part of the app that presents the intended use. Even if a software has an IFU, there may still be benefit in embedding the model card info into the app itself so users find and/or see it when relevant. 

Is FDA guidance still in draft form? And if so, what does this mean? 

Yes, the guidance is still draft. We have not heard about if/when it will be finalized. However, draft guidance generally reflects FDA's current way of thinking about a topic. The process described in draft guidance can evolve as the FDA receives feedback on the draft and as they review more submissions for which the draft is applicable. As such, it is beneficial for manufacturers to be aware of the draft guidance and consider how your HF (and other) processes can address some of the concerns the FDA articulates in the guidance. Because the guidance (and AI-enabled medical devices generally) are so new, we encourage you to communicate with the FDA before any submission to confirm that your planned approach aligns with their latest expectations.  

Contact our team to learn more about integrating human factors engineering into the development of AI-enabled products.  

Erin Davis is a Research Director, Tess Forton is a Managing User Researcher, and Julee Henry is a Lead Human Factors Specialist.