Ensuring AI transparency and accountability is critical for building trust and safely and effectively implementing AI systems in health. Given the inherent resistance to change, including inertia and conservatism, in the medical field, establishing transparent regulatory frameworks is essential. This involves developing nuanced regulatory approaches that keep pace with rapid AI advances, as traditional one-size-fits-all regulation is inadequate for the diverse and evolving nature of AI technologies. Creating adaptable frameworks that address the specific characteristics of different AI technologies is crucial. Harmonizing data- sharing policies across borders facilitates global collaboration while maintaining security and privacy, which is essential for the widespread adoption of AI.Maintaining human oversight of AI systems is of paramount importance for safeguarding trust, efficacy and ethical standards. The World Health Organization (WHO) emphasizes that the “principle of autonomy requires that the use of AI or other computational systems does not undermine human autonomy. In the context of health care, this means that humans should remain in control of health- care systems and medical decisions.”17 Given the complexities of AI, especially genAI, human involvement is necessary to validate AI outputs and support decision-making. If a culture of trust and collaboration is established, ensuring that AI systems are safe, reliable and accepted by all stakeholders, AI can be effectively integrated into health provision.3.3 Low confidence in AI within a fragmented regulatory and governance framework Fragmented, outdated regulations that hinder AI innovation Tactically, AI-driven health faces four significant regulatory challenges in the short term. First, there is a fragmented regulatory landscape with a divide between countries with stringent regulations and those without.18 Second, the perceived lack of regulatory clarity, where regulations do not keep pace with the advance of AI, stifles innovation. Third, the regulation of software and AI uses a one- size-fits-all approach that might not be fully relevant for genAI. Finally, AI development and regulation without data-sharing rules remains a challenge. The gap is growing between countries leading the race in AI regulation and those without the means to engage in this new field. Public engagement and political will are essential: in the US, the 2023 Executive Order on the Safe, Secure and Trustworthy Development and Use of Artificial Intelligence19 was one catalyst for progress. Supporting emerging economies and local or regional regulation approaches will help mitigate inequities: this should remain part of international development priorities as well as national priorities. Slow regulation often stems from the expectation that AI must be fully developed before implementation, and that risks should be mitigated with ambitious regulations. This conception is almost ineffective by design for two reasons: it does not use all of the tools available for mitigating risks, and it does not address a core problem of AI, which is the regulation and control workload. On the first point, AI risk mitigation should use the full portfolio of tools: guidance and standards in particular are more flexible than regulation, and they still help to strike the right benefit-to-risk ratio. Giving more space and more funding to support early guidance and standards could help the ecosystem avoid premature regulation that would hinder innovation. The Future of AI-Enabled Health: Leading the Way 20

The Future of AI Enabled Health 2025