The government-led AI Summit Series launched at Bletchley Park in England in 2023 has produced an increasingly elaborate, well-attended and genuinely ambitious process for governing one of the most consequential technologies in human history. It has not yet produced a single binding commitment.

Nevertheless, to say the Summits have produced nothing would be unfair. 

The Summits are the most sustained multilateral effort to govern a general-purpose technology before its harms become irreversible. Early iterations of the Summit led to AI Safety Institutes. They led to non-binding uses of safety commitments from frontier companies or highly capable general-purpose AI models and systems. They also led to an international scientific report on AI risks and capabilities. 

When the dominant powers disengage from multilateral cooperation, the countries in between gain negotiating weight they rarely hold.

The ambition to govern AI from such an early stage deserves acknowledgment and scrutiny in equal measure. AI has become, by every meaningful measure, more powerful and less governed today than when governments first gathered at Bletchley Park in 2023. 

The relentless power of AI matters urgently because the country controlling roughly three-quarters of global AI compute capacity is no longer meaningfully inside the governance architecture. 

The United States declined to sign the Paris Statement on Inclusive and Sustainable Artificial Intelligence for People and the Planet in 2025, which called for open, inclusive and ethical approaches to AI. 

In Delhi, the U.S. explicitly rejected multilateral governance while endorsing the non-binding Delhi Declaration, which committed to international cooperation and multistakeholder cooperation. The U.S. signed this declaration simply because it was voluntary without enforcement mechanisms. The 2027 Summit in Geneva cannot afford to make the same mistake.

The absence of the U.S., however, is not necessarily a problem. It may be an opening. 

When the dominant powers disengage from multilateral cooperation, the countries in between gain negotiating weight they rarely hold. Middle powers such as France, Germany, Finland, Spain, Italy, India, Brazil, Japan, Kenya, Canada and the United Kingdom have more leverage in Geneva than at any point in this process.  

In India, billions of people encounter AI not as a frontier model but rather as a possible welfare system, an agricultural platform or a biometric database. This distinction matters.

Recognizing the power of middle powers matters because in July, the UN’s Global Dialogue on AI Governance will hold its first full session in Geneva. Decisions in Geneva will determine whether the convergence of the two AI-focused convenings yields tangible governance mechanisms and fills capacity gaps or results in yet another declaration. 

These two meetings converge in Geneva and hold enormous consequences for AI governance globally. 

The AI Summit Series: From Safety to Society — and Back?

There are clear messages that we can glean from changing summit titles — from Safety (2023) to Action (2025) to Impact (2026) — that reflect a broader shift in focus away from AI safety and governance and toward societal impact. 

The shift from safety to impact is understandable. The series in India, for example, was the first time the meetings were hosted by a country in the Global South. India’s role in hosting the meeting was not a symbolic gesture but a structural change that reflected whose priorities shaped the global agenda. In India, billions of people encounter AI not as a frontier model but rather as a possible welfare system, an agricultural platform or a biometric database. This distinction matters.

India signaled this shift away from safety as the organizing frame toward diffusion, adoption and development outcomes. However, that shift has amplified industry-focused national AI ambitions. The Delhi Summit amplified investment pledges, CEO roundtables and 500,000 visitors through an exhibition floor.

This focal transition from safety to profit raises a question Geneva must answer directly: If the AI Summit Series now covers AI and society in its broadest sense, what is its specific governance function? 

If the answer is that governance needs its own dedicated forum, then the UN Global Dialogue on AI Governance is precisely that, and the two tracks need to stop running in parallel and start working as a system.

Meaningful participation in AI governance requires the capacity to participate in AI development. The Global Fund for AI Capacity Development, if designed correctly, is not charity but the only meaningful measure of inclusion.

One way to resolve this tension is through structural change. The process should explicitly differentiate between the two tracks:

• Track 1: Frontier AI safety, focused on advanced systems, risk thresholds and company accountability through independent evaluation
• Track 2: Development and diffusion, focused on access, capability-building, infrastructure, and equitable distribution of benefits

This is not about exclusion. It is about functional clarity. Safety requires deep technical expertise. Development requires broad participation and political buy-in. Trying to do both in a single track risks weakening both.

If governance is to remain credible, these tracks cannot continue to evolve in parallel without coordination. They must be designed to function as a system.

The UN Global Dialogue on AI: Built for Everyone, Moving for No One

In 2024, 193 countries signed the UN’s Global Digital Compact. 

Against the summit series’ ten-country Safety Institute Network, that number captures the gap between the two tracks precisely — the UN remains the only forum where smaller states have both a voice and a vote, but inclusion alone does not guarantee participation. There is no broad, globally distributed expertise on frontier AI risk. Many states lack the capacity to engage meaningfully in technical governance debates. 

Getting 193 countries to agree on anything is, by design, slow. That is the constraint built into the architecture of multilateral collaboration.

What the UN track has built is structurally promising. The Independent International Scientific Panel on AI is composed of forty experts from around the world and modeled on the Intergovernmental Panel on Climate Change (IPCC). The panel is designed to produce assessments that inform governance across political cycles. 

The Global Dialogue, launched in September 2025, is the first platform where countries excluded from G7, G20 and OECD processes have equal procedural standing. 

AI governance needs an architecture that functions as a system capable of uplifting countries with limited access to computing.

The UN secretary-general has proposed a Global Fund for AI Capacity Development, an international financial mechanism targeting between $1 and $3 billion, to bridge the widening digital divide. The primary objective of the fund is to ensure a fundamental minimum capacity for every nation, empowering economically developing countries to participate in and benefit from the burgeoning AI economy. This foundational capacity is defined by four pillars: cultivating technical skills to foster local expertise; providing affordable compute  (or raw processing power, hardware, and software); developing representative data, and increasing access to advanced AI models.  

According to the International Telecommunication Union (ITU), 2.2 billion people remained offline in 2025, with 96% of the people offline residing in low- and middle-income countries. Africa, home to 18% of the world’s people, accounts for less than 1% of global data center capacity. In Kenya and Senegal, a single graphics processing unit costs 75% and 69% of GDP per capita, respectively. These units are essential for high-performance computing, making it nearly impossible for these countries to keep up with countries that have a larger proportion of their population online. 

Meaningful participation in AI governance requires the capacity to participate in AI development. 

The Global Fund for AI Capacity Development, if designed correctly, is not charity but the only meaningful measure of inclusion.

What Geneva Must Deliver

Geneva is the first moment both tracks occupy the same city. This setting leverages Switzerland’s functional neutrality as one of the few diplomatic settings where geopolitical tensions can be partially suspended in favor of process.

For the UN Global Dialogue in 2026 and the Geneva Summit in 2027 to deliver credible AI governance, four things must change.

First, the Scientific Panel’s findings must be integrated as the evidentiary basis for specific outcomes and linked to governance. A procedural link between science and commitment must be built in Geneva before the 2027 agenda is set. Without it, the Panel becomes the most credible research body in AI governance that no one is required to act on.

Second, the principle on which AI governance is set and enforced is essential. Enforceable commitments with a compliance mechanism are essential, whether it be compute access, pre-deployment safety standards, or AI in armed conflict. 

The upcoming events in Geneva must elevate the collective voices of middle powers to enact global AI governance that focuses on not only innovation but also safety and equity.

One credible precedent matters more than a broad but unenforceable declaration. It establishes that governance is possible in practice, not just in principle. The test is simple: Does the commitment change behavior, or only signal intent?

Third, coalitions among middle powers are forming in trade forums, in digital infrastructure partnerships and in regional AI strategies. Geneva is the first moment those alignments can be formally brought into the global AI governance architecture. The current moment requires that middle powers engage across blocs rather than choose between them. 

Fourth, the proposed Global Fund for AI is the most concrete structural initiative the UN has put on the table. It will be between $1 and $3 billion available for low-income countries, and its design will determine whether it reduces dependency or reproduces it. By enhancing access to computing, open-source model development, and local-language AI, countries may develop the standing to shape the rules. 

AI governance needs an architecture that functions as a system capable of uplifting countries with limited access to computing.  Fragmentation between AI haves and have-nots is already underway. Capacity is not separate from governance. It determines who can shape the rules. 

The upcoming events in Geneva must elevate the collective voices of middle powers to enact global AI governance that focuses on not only innovation but also safety and equity.