The European regulatory ecosystem is undergoing structural transformation. The AI Act, MDR/IVDR, and EU data regulations (including the EHDS Act) are converging to form an integral legal architecture that redefines what it means for digital health technologies to be trustworthy, effective and legally viable.

This session explores how digital health innovators can align clinical evidence, algorithmic design and data governance across these frameworks without fragmenting compliance or stalling innovation. As of AI, device, and data law continue to converge, strategic legal alignment becomes essential to ensure that products meet both regulatory expectations and the complexity of real world use.

The integration of artificial intelligence (AI) into healthcare presents unique regulatory challenges, from managing adaptive algorithms to addressing risks like automation bias and model drift. Traditional regulatory frameworks, designed for static technologies, often struggle to accommodate the complexity and dynamism of AI as a medical device (AIaMD). In response, sandboxes offer a novel, collaborative environment for AI developers, regulators, and clinical stakeholders to test and refine regulatory approaches in various environments (theoretical and real world). This panel contribution shares key insights from the AI Airlock pilot, which supports selected AIaMD developers through tailored regulatory experimentation [and explores the EU’s approach – TBC].

Using a regulatory challenge-led framework, the AI Airlock pilot explored issues across the product lifecycle, including AI errors, validation, explainability, performance drift, and human-AI interaction. Now in its second phase, it continues to explore some of the most pressing regulatory challenges with AIaMD today. The session will discuss how sandboxes can support regulatory learning while protecting patient safety, foster evidence generation for novel technologies, and inform future policy. Reflections will also cover practical lessons on setting up a sandbox and cross-sector collaboration.

The intended purpose of a device goes beyond regulatory concerns; it also drives health economic considerations and can underlie intellectual property issues, all of which inform market access strategy. Aligning the intended purpose across all domains is crucial for successful market entry.

.

This session will explore how AI agents can help regulatory and clinical teams reduce the time and effort needed to produce compliant clinical evaluations under MDR. Drawing from real-world examples, I’ll share how automation and intelligent workflows can simplify article screening, data extraction, and evidence appraisal, without compromising quality. Attendees will walk away with practical tools, use cases, and a realistic view of what AI can do today in the context of clinical evaluation.

Fulfilling its intended use, whilst being safe for that intended use is the basic requirement for all medical devices. The increase in the availability and use of artificial intelligence in medical devices comes alongside the publication of new regulations and standards to help control and guide manufacturers producing these devices. ISO 14971:2019 is a high-level, process standard that describes risk management for all medical devices. It does not provide guidance on how to apply its requirements to different types of medical devices, nor should it. We have some support from BS/AAMI 34971:2023, providing guidance to industry on how to apply ISO 14971:2019 to machine learning-enabled medical devices, but how should it be implemented?

As medical devices are seen to become more complicated with the inclusion of artificial intelligence and machine learning elements, there could be a temptation to think that our risk management process must evolve into something novel and even more complicated compared to our older, less advanced medical devices. That is not essentially the case.

There are concerns about transparency, explainability and bias. Rightly so. There will always be uncertainty about novel technologies, which are thus seen as high-risk and worrying.

New technologies must mean new risks and higher risks. Maybe, maybe not.

Does all of this mean that our risk management process needs to drastically evolve? Not so much.

This session will:

• Look at how a risk management process can evolve with newer technologies;
• Advocate for the foundations of the risk analysis that are critical to risk management of all medical devices;
• Discuss basic risk analysis methods that are critical to identifying potential risks for medical devices enabled with machine learning;
• Improve understanding of where risk controls will be implemented in the development of machine learning-enabled medical devices;
• Highlight the connection between risk, data management, usability and clinical evidence;
• Suggest post-market surveillance activities to be considered for machine learning-enabled medical devices.

This presentation explores the clinical evaluation requirements for Software as a Medical Device (SaMD) under the EU Medical Device Regulation (MDR), with a particular focus on the use of clinical evidence through equivalence and the application of Article 61(10). It outlines the regulatory criteria for establishing equivalence for SaMD, highlights the challenges associated with demonstrating sufficient clinical evidence in the absence of clinical investigations, and clarifies the conditions under which Article 61(10) may be applied. Practical examples will be used to illustrate key principles and support consistent interpretation of the MDR requirements.