Earning Trust for AI in Health 2025

The need for a pragmatic approach: Guidelines, sandboxes and post-market surveillance 2 The most effective way forward for AI innovation in healthcare combines regulation with post-market performance monitoring. Globally, governments are actively prioritizing the development and updating of legislation related to data protection and AI. This regulatory momentum reflects a global recognition of the need to manage the implications of AI technologies: “policymakers have progressed from the ‘understand’ stage … to the ‘shape’ stage”.19 However, AI regulation remains nascent across regions with variable balance between allowing innovation and enforcing regulation and security. Legislative developments are often slow and are not easy to adapt in the face of the rapidly changing AI environment. AI is still an emerging technology, for which opportunities for new applications are regularly discovered. Legislative developments pertaining to AI technologies will need to ensure that any novel developments are not stifled. For instance, the EU AI Act only outright forbids the use of AI for certain purposes and practices that are inconsistent with the norms and values of the EU.20 Beyond that, it indicates certain areas (e.g. medical devices) where additional scrutiny is warranted yet imposes no limitations on how AI can be deployed within those areas, thus safeguarding the innovation potential. National legislative initiatives may lead to a fragmented international landscape on topics such as AI standards, sharing of best practices or mutual recognition of regulations. Navigating changing legislative environments can result in short-term uncertainty that may temporarily hamper innovation, as companies may be hesitant about investing in new ideas until the regulatory context becomes clearer. Fragmentation also makes it challenging for AI technologies to be scaled within regions, as the market access requirements in different countries within a single region may not align. On a global level, a degree of fragmentation is to be expected as different norms and values informing market access processes underpin health systems worldwide.21 Nonetheless, multilateral cooperation could drive increasing regulatory convergence over time. There is a strong case for a guidelines- based approach to complement legislative frameworks and establish more nuanced and detailed provisions. Regulators can continue to build on existing practices22 such as those used to permit certain medical products before full market authorization23 (see Box 1). These practices are governed by regulatory frameworks and operationalized through guidelines, thus making use of the flexibilities that guidelines provide while ensuring that any innovation aligns with societally acceptable boundaries.2.1 Legislation can build a strong baseline for governing AI in health Earning Trust for AI in Health: A Collaborative Path Forward 10