Regulatory strategies for AI-assisted software
Medical devices incorporating AI-enabled technologies raise unique regulatory questions due to their iterative and potentially self-updating nature, which is incongruent with historical regulatory approaches. The use of machine or deep learning offers the opportunity for continual optimization of an algorithm as new training data becomes available; whereas traditional regulatory approaches have focused on frozen algorithms that require new clearance or approval if changes are made over the life of the commercialized software.
While the global regulatory landscape varies, one threshold question with which stakeholders must grapple is whether an AI-assisted software qualifies as a medical device in the jurisdiction of interest. If software that embeds AI features meets the definition of a ‘medical device’ according to local regulations, then the AI-assisted software would have to comply with general safety and performance requirements established for such devices.
Adding to the complexities of this already highly regulated area, the landmark EU AI Act now in force will impose an extensive set of new obligations on companies across all sectors with broad extra-territorial scope. Assessment of medical devices under the AI Act will add new (and in some ways overlapping) assessment obligations to the existing regulations for devices. Medical device manufacturers and other stakeholders will need to align with each of these standards, which may be challenging in balancing safe and responsible AI without having a negative impact on innovation.
Additionally, while the U.S. federal government has yet to adopt far reaching laws or regulations that govern the use of AI, the states, with Colorado being the first, are enacting laws that may result in a patchwork of obligations across the states. In the coming months, we expect that other geographies will also adopt laws and regulations that govern the use of AI more generally as well as specifically AI used in health care.