¹ú²úÊÓƵ

In May 2026, ¹ú²úÊÓƵ’s Technology & Democracy programs convened the inaugural dialogue of the Shangri-La Series: AI for Middle Powers at the Doris Duke Shangri La Center for Culture and Ideas in Honolulu, HawaiÊ»i. Over two days, 21 senior policymakers, technologists, civil society leaders, and researchers from across Africa, Asia, Latin America, North America, and Europe gathered for a closed-door conversation about one of the defining questions of our time—who gets to shape the rules of artificial intelligence, and on whose terms?

What Is the Shangri-La Series?

The Shangri-La Series is a multi-year convening series organized by ¹ú²úÊÓƵ’s Technology and Democracy Programs. The Series is built on a core premise that the global conversation about AI governance has been dominated by a small number of large technology companies and wealthy governments—namely the U.S. and China. All the while, the global majority has had limited say in defining the norms, standards, and institutions that will govern this technology for decades. The Shangri-La Series exists to define alternative paths. 

What We Discussed

The Honolulu dialogue brought together leaders from what are often called “middle power” countries—an intentionally broad term that, for our context, which included nations that are neither the dominant AI developers nor passive bystanders, but active shapers of how AI is adopted, regulated, and governed within their borders and in international forums. The conversations were structured as an off-the-record exchange, allowing participants to speak candidly about the pressures, trade-offs, and opportunities they face. Here are some small highlights of the themes and tensions that emerged over the three-day convening.

The Opportunity Space – Where can middle powers act? A central insight from the convening was that countries that don’t build frontier AI models can still exercise meaningful influence by shaping how those models are adapted, evaluated, and deployed. The “adaptation layer”—how AI is customized for local languages, laws, institutional contexts, and values—is where middle powers have the most leverage, and an area for deeper exploration within global governance conversations. 

Evaluation is an underrated lever. Participants articulated that middle powers cannot be left only to manage AI’s downstream effects, and that they must have a voice in the larger questions around safety and evaluation. The absence of a robust, independent AI evaluation infrastructure outside global North institutions of higher education was a frequently noted critical gap. Interest emerged in defining and enforcing meaningful standards for how AI systems perform, not just across technological benchmarks but also societal ones. The notion of assurance gained traction across the group, and there was agreement that working in coalition, individual nations can leverage their distinct technological advantages and market size to shape market behavior and international norms. 

Dependency is a real risk. There was frank discussion of the ways AI supply chains, cloud infrastructure, and data flows can concentrate economic and political power in the hands of a small number of actors. Participants acknowledged the challenges posed by uncritical adoption and by more isolationist “AI sovereignty”, participants explored multiple elements that could constitute third paths and generate leverage—technological paths, like shaping the application layer, leveraging existing DPI infrastructure, and developing open, low-resource task-specific AI; policy paths, like developing a middle power policy stack and codifying evaluation standards; and geopolitical paths, forming coalitions and setting norms.

Values cannot be deferred. Across sessions, participants returned to the question of whose values get embedded in AI systems. Values are not an afterthought—they are already being codified into models by the small number of powerful players who are building them. We see this at both the application level, around elements like culture, language, and local context as well as at a more foundational level around democracy, safety, and openness. Middle powers cannot be left only to manage the former, and the field must develop adequate tools for middle powers to exert a meaningful voice for both.

What Comes Next

The May 2026 Honolulu dialogue was the first of two planned convenings in the Series. A landscape analysis of deployments across the global majority will be released in early July 2026, and a comprehensive synthesis report capturing the key themes and insights from the Hawaii dialogue is currently in development. A second dialogue is planned for 2027, building on the relationships, analysis, and open questions from Honolulu.

We will also seek to ground the Series’ work in a specific regional context, and the team is exploring opportunities to bring the Shangri-La Series into dialogue with other international AI governance processes.