The UN Global Dialogue in Geneva, held from 6-7 July, brought together governments, companies, academics, and civil society to discuss the risks emerging from the widening technological gap between middle powers, developing countries, and the AI superpowers.
Advanced semiconductors, computing infrastructure, and AI capabilities remain concentrated primarily in the United States and China, leaving many countries dependent on technologies they neither develop nor fully control.
This creates difficult choices between proprietary US systems and open-weight Chinese models, while raising issues around safety, governance, and technical capacity. The resulting asymmetry extends beyond access to AI, limiting the ability of other countries to shape the standards, infrastructure, and governance arrangements that will determine AI’s future.
The UN summit resulted in a proposal for a ‘Dialogue of Dialogues’ approach to help address some of the issues discussed. This approach cannot counter the technological dominance of the US and China. But it can help countries strengthen their collective position by linking existing governance initiatives, improving technical interoperability, developing shared safety baselines, and building greater capacity to shape the future of AI governance.
Risks and governance priorities
The Geneva summit drew on the preliminary report of the Independent International Scientific Panel on AI. The report identifies various real-world harms emerging from artificial intelligence, including the lack of robust mechanisms to evaluate autonomous AI agents and mitigate bias; the proliferation of AI-generated child sexual abuse material and deepfake-enabled sexual violence; and the environmental cost of building large-scale AI infrastructure.
Notably, the report also describes how AI governance initiatives remain fragmented, with over 40 types of AI governance instruments globally. It also points out that governance is largely voluntary and concentrated at the corporate level.
Another report by the AI Governance for Humanity Lab takes stock of this fragmented landscape and proposes concrete steps for streamlining it through interoperable standards, like shared incident reporting and crisis protocols.
The Geneva summit built on this work by proposing a ‘Dialogue of Dialogues’ approach, resembling the IPCC, ICAO, and WHO models. This would connect governance initiatives, strengthen interoperability, and support evidence-based assessment without creating a supranational regulator.
Through the UN AI Dialogue Partnerships Hub, it seeks to foster consensus, knowledge-sharing, and collaboration among like-minded state and non-state actors while respecting geopolitical diversity and national sovereignty.
Coalitions to ‘govern the governance’
The Dialogue of Dialogues would enable smaller coalitions to advance shared AI priorities, preventing gridlock and aiding broader multilateral progress. This includes minilateral and regional efforts among states, and multistakeholder efforts uniting like-minded governments, tech companies, and civil society.
One good format to build on would be compute coalitions, where groups of interested countries pool their computing resources, infrastructure, and expertise to reduce their dependence on the AI superpowers. The EU’s AI Continent Action Plan is an obvious example. So too is Brazil’s Artificial Intelligence Plan (PBIA), which supports 100 collaborative AI research projects with Latin American and African partners. Meanwhile, Africa’s AI factory initiative promotes digital sovereignty by developing locally relevant AI models in African languages and pricing services in local currencies.
Another useful format would be regulatory coalitions grouped around shared challenges. This would see a coalition of governments, researchers, civil society, and AI laboratories identify practical common standards on challenges such as AI watermarking, deepfake detection, or safeguards against AI-generated sexual exploitation material.
Such standards could form the basis for adoption by others: the European Union’s AI Act is a good example, as is Singapore’s proposed ASEAN Digital Economy Framework Agreement (DEFA) which aims to help establish common rules on digital trade and trusted cross-border data flows.
Smaller coalitions like these could promote technical interoperability through common evaluation methods. That would enable AI developers, particularly in technologically underdeveloped countries, to demonstrate compliance across regulatory regimes without costly duplication of certification. Such interoperability could significantly reduce compliance costs for companies while enabling them to compete internationally.
What a Dialogue of Dialogues can achieve – and what it can’t
The UN’s Dialogue of Dialogues cannot impose binding constraints on the US or China, or somehow compete with the US-led Pax Silica or China’s WAICO groupings. It will be most effective in promoting technical interoperability and shared baselines among other countries – creating ‘translation keys’ that allow diverse AI regulatory regimes to communicate with each other.
The risk is that, if handled incorrectly, it could lead to further fragmentation. Regulators must work with the private sector to promote interoperability: for example, by developing shared safety benchmarks or transparency reporting requirements.
Interoperability is not a cure-all for the polarizing dynamics of the global AI race. But it can reduce friction among countries not locked in strategic competition, ensuring that countries do not have to choose between incompatible regulatory regimes such as the US and China or the EU.
Such an effort could start with what is achievable. For example, sovereign coalitions could seek to create a ‘minimum viable interoperability’ by 2027 – as suggested by Costa Rica in the Geneva Dialogue. This could be achieved by establishing shared terminology, comparable risk classifications, consistent documentation, interoperable incident reporting, and multilingual evaluation methods. These are technical, not overtly political measures and can be built even amid rising geopolitical tension.
The UK is well-placed to be a key player in global AI governance interoperability by rallying support among middle powers. It has a strong AI safety research base, is a signatory to Pax Silica, and maintains dialogue channels with both the US and China as well with technologically developed and underdeveloped countries through the Commonwealth.
The Dialogue of Dialogue approach will also need to be defended at the United Nations. The Global Dialogue was established by General Assembly resolution, not by Secretary-General prerogative so a new UN Secretary General cannot unilaterally abolish it. But there is a risk it could be deprioritized.
To ensure it has lasting impact, the Dialogue of Dialogues should be progressed by the UN’s Office for Digital and Emerging Technologies (ODET), which oversees the UN’s 2024 Global Digital Compact and emerging technology governance. ODET is best placed to coordinate across the UN system and ensure balanced participation among governments, civil society, academia, and industry.
Co-locating with the ITU’s AI for Good and WSIS summits would provide an ideal opportunity to promote the Dialogue of Dialogues approach to industry and a wider range of stakeholders, while reducing resource requirements and leveraging an established audience. This arrangement, however, should be accompanied by robust independent governance safeguards under ODET to prevent undue corporate influence.
The Dialogue of Dialogues approach cannot prevent the challenges that countries face in the emerging AI landscape. But it should be persevered with, if countries outside of great power rivalry are to gain equal access to the opportunities AI promises – and promote the standards needed to ensure AI uses are safe and beneficial to all.
Ultimately, bridging the global AI divide requires sustained, inclusive multilateralism. Giving every nation a voice in setting global standards through the UN can ensure that AI serves as a tool for universal progress rather than further fragmentation.