The integration of artificial intelligence (AI) technologies into clinical practice stands as a defining challenge and opportunity for global health in the coming decade. While use cases such as clinical documentation and diagnostic support are becoming increasingly common in high-resource settings, where over half of clinicians now report using AI tools, much less is known about frontline clinicians’ experiences, barriers, and real-world needs in low- and middle-income countries (LMICs).

In LMIC contexts, large language model (LLM)-enabled clinical decision support systems (CDSSs) offer the potential to extend high-quality decision support to settings with limited human and technical resources. Yet they also carry significant risks, including language bias, inequitable access, limited local capacity for oversight, and the possibility of recommendations that are misaligned with local medical guidelines that undermine patient safety or trust.

Over the past two years, rapid advances in generative AI have outpaced the establishment of evidence, governance, and capacity frameworks required for responsible use. Health systems and policymakers are now faced with urgent questions:

• How can AI tools be evaluated for clinical safety and contextual relevance?
• What data infrastructures and regulatory mechanisms are needed for equitable deployment?
• And how can local researchers and health institutions lead, not just participate, in AI innovation?

These questions form the basis of this CDSS Learning Agenda, which aims to generate collective insight and actionable evidence to guide safe, effective, and equitable integration of AI into health systems. The agenda identifies six thematic priorities—localization and language equity; evaluation and real-world evidence; voice-enabled and multimodal tools; capacity and local ownership; governance and trust; and infrastructure and enabling environments—that collectively define the ecosystem required for sustainable AI adoption in LMICs.