Global Dialogue highlights need for interoperable AI governance

Building safe, secure and trustworthy AI requires countries to align their governance frameworks rather than adopt a single global regulatory model, participants heard on the second day of the UN Global Dialogue on AI Governance. Speakers from governments, international organisations, industry and civil society argued that interoperability, backed by common standards, scientific evidence and inclusive participation, is essential to address AI risks that increasingly cross national borders.

The discussion also highlighted a growing imbalance in global AI development, with participants warning that governance should not be shaped solely by the countries and companies leading frontier AI. Instead, they called for developing countries to become co-creators of international AI governance through stronger capacity development, shared standards and multilateral cooperation.

AI concentration risks becoming governance concentration

Opening the session, co-chair Paula Bogantes Zamora, Costa Rica’s Minister of Science, Innovation, Technology and Telecommunications, argued that the world has reached a point where agreeing on AI principles is no longer enough.

‘The world does not need more AI principles, it needs a common way to prove they’re being implemented.’

Bogantes Zamora warned that AI development remains heavily concentrated. She noted that institutions in the United States produced 59 notable AI models in 2025 and China another 35, while the rest of the world produced just 13. She argued that this concentration of infrastructure also creates a concentration of evidence, allowing a small number of actors to determine which risks are measured, which benchmarks are accepted and how AI safety is evaluated.

She also pointed to findings showing that 118 countries, primarily in the Global South, remain largely absent from major international AI governance discussions.

Rather than pursuing regulatory uniformity, Bogantes Zamora proposed what she called ‘minimal viable interoperability’ by 2027, including shared terminology, comparable risk classifications, interoperable incident reporting and multilingual evaluation methods that allow different governance systems to function together.

Interoperability should connect governance systems, not replace them

Co-chair Rebecca Finlay, CEO of the Partnership on AI, argued that governance efforts must be grounded in stronger scientific evidence and greater transparency.

She outlined three priorities: strengthening independent scientific research, improving public access to evidence through greater disclosure by AI developers, and creating shared baselines for measuring progress in the public interest.

‘The panel provides the evidence and the dialogue provides the direction,’ Finlay said, describing the UN scientific panel and the Global Dialogue as complementary processes.

UN Under-Secretary-General and Special Envoy for Digital and Emerging Technologies Amandeep Singh Gill echoed that message, warning that fragmented AI governance creates regulatory arbitrage, accountability gaps and unnecessary compliance burdens, particularly for smaller companies and developing countries.

Rather than harmonising all AI rules into a single global framework, Singh Gill argued that countries should focus on building practical bridges between different governance approaches.

He also highlighted the emergence of increasingly autonomous agentic AI systems as a new governance challenge requiring adaptive oversight mechanisms, including cross-border regulatory sandboxes and continuously updated risk assessment frameworks.

Existing frameworks provide building blocks

During the first panel, speakers pointed to several initiatives that could serve as foundations for greater interoperability.

Yoichi Iida, adviser at Japan’s Ministry of Economy, Trade and Industry, highlighted the OECD AI Principles and the Hiroshima AI Process as examples of frameworks already helping countries align governance approaches despite different legal systems.

Syed Ahmed of Infosys said that translating broad principles into practical implementation remains technically challenging.

Using transparency as an example, he explained that the concept carries different technical requirements across governance frameworks, requiring detailed mapping of individual controls rather than simply aligning high-level principles.

Nouf Al Hameli of the UAE Ministry of Foreign Affairs similarly argued that countries define concepts such as ‘high-risk AI’ in different ways, making common incident reporting and mutual recognition of governance practices increasingly important.

Leonardo Cervera Navas, Secretary-General of the European Data Protection Supervisor, compared AI governance to aviation safety, arguing that while countries operate different legal systems, they nevertheless follow common international safety rules.

‘The higher the risk, the higher the care and supervision required,’ he said, referring to the EU AI Act’s risk-based approach.

Inclusive evaluation and trustworthy evidence remain critical

Several speakers argued that trustworthy AI depends not only on technical standards but also on ensuring that governance reflects linguistic, cultural and demographic diversity.

Dr Joy Buolamwini, founder of the Algorithmic Justice League, warned that widely used AI benchmarks often fail to represent the global majority, noting that some have historically included less than 5% of the world’s population.

She called for harm reporting systems that record not only technical failures but also who was affected, creating stronger foundations for accountability and redress.

Celeste Saulo, Secretary-General of the World Meteorological Organization, drew lessons from more than 150 years of international weather cooperation, arguing that trust cannot simply be declared.

‘Trust must be built through verification,’ she said, pointing to the organisation’s longstanding use of shared standards and independent validation across 193 countries.

Qinghua Lu of Australia’s CSIRO proposed greater collaboration through shared evaluation methods, common risk management principles and international testing exercises that include multiple languages and national contexts.

Global South calls for a stronger role in shaping AI governance

Interventions from member states and stakeholders repeatedly stressed that interoperability should not become another mechanism for exporting governance models developed elsewhere.

Pakistan argued that AI safety standards are currently shaped by a small group of countries and companies, calling instead for genuinely multilateral governance under the UN.

Brazil similarly stressed that interoperability must not undermine digital sovereignty, while South Africa argued that governance frameworks should reflect the realities of developing countries and support technology transfer and capacity development.

Other speakers highlighted practical priorities, including multilingual benchmarks, common standards for documenting AI training data, cross-border incident reporting systems and greater participation from local governments, academia and civil society.

Concluding the discussion, both co-chairs argued that trustworthy AI depends not on identical regulations but on governance systems that can communicate, exchange evidence and recognise one another’s safeguards.

They identified shared technical standards, independent evaluation, multilingual benchmarks, human rights protections and continuous multistakeholder cooperation as the foundations for AI governance capable of working across borders, while warning that progress will depend on maintaining momentum between international meetings rather than restarting discussions each year.

Track all key moments from the Global Dialogue on AI Governance inaugural meeting on our dedicated page.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!