IEEE

Governing AI in a Fragmented World: Multilateral Frameworks, Geopolitical Rivalries, and the Stakes for the Global South

Transparency, accountability, and human oversight;Interoperability of governance approaches;Open-source software, open data and open AI models;AI capacity-building;

Governance in a Geopolitically Fractured World The central crisis in global AI governance is not technical - it is structural. The world's leading AI powers are operating under fundamentally incompatible frameworks, and the Global Dialogue risks becoming irrelevant if it fails to address this fracture directly. Interoperability of governance approaches is the load-bearing pillar. The US, EU, and China are each exporting distinct governance models - market-led, rights-based, and state-controlled respectively. Without deliberate interoperability mechanisms, the result is not three systems coexisting - it is fragmentation that forces every other nation to pick a side. This is the structural problem. Everything else is downstream of it. Open-source AI models and data are where the geopolitical competition is currently being won quietly. China's strategic deployment of open-source models like DeepSeek across the Global South is not a technical decision - it is a governance play. Free, deployable AI builds dependency and allegiance faster than any bilateral agreement. The Dialogue must recognize open source not merely as an innovation tool but as a vector of geopolitical influence that demands governance attention. AI capacity-building is the arena where the US-China rivalry is being actively contested. Developing nations are not passive observers - they are the prize. Whichever power funds infrastructure, trains talent, and deploys AI solutions in these markets will shape governance preferences for a generation. Ignoring capacity-building means ignoring where the real decisions are being made. Transparency, accountability, and human oversight become critical precisely because the most powerful AI actor has stepped back from multilateral commitments. Accountability frameworks without US participation are structurally weakened. Naming this gap - and building mechanisms resilient enough to function without universal buy-in - is the most honest and urgent policy challenge before this Dialogue. Together, these four priorities form a coherent, power-aware governance agenda. Not aspirational. Actionable.

The United States stands at a paradoxical inflection point. As the world's leading AI innovation economy, it has the most to gain from effective global governance — and currently the most to lose from its absence. On Interoperability, the proliferation of incompatible regulatory regimes directly threatens US technology exports and market access. American companies operating across jurisdictions face compounding compliance burdens as the EU AI Act, China's algorithmic regulations, and emerging national frameworks diverge rather than converge. Without US engagement in interoperability mechanisms, American industry will be forced to comply with rules it had no hand in writing — a strategic and commercial liability. On Open-Source AI, the United States faces a concrete competitive threat. China's deliberate deployment of open-source models across developing markets is eroding US technological influence in regions where American companies could otherwise lead. The absence of a US-backed open governance framework for open-source AI creates a vacuum that is being filled — systematically and strategically — by others. On Capacity-Building, US bilateral AI partnerships with Saudi Arabia and the UAE represent opportunity, but bilateralism alone cannot substitute for multilateral standard-setting. Nations receiving capacity-building support from alternative providers will adopt governance norms aligned with those providers. The United States risks winning individual deals while losing the broader normative landscape. On Transparency and Accountability, US withdrawal from multilateral frameworks does not eliminate accountability demands domestically. Congress, civil society, and the courts are generating governance pressure regardless. Engaging internationally would allow the US to shape accountability standards proactively rather than reactively absorbing frameworks designed without its input. The opportunity is clear: re-engagement now, on US terms, before the rules are finalised without Washington at the table.

The Global Dialogue on AI Governance arrives at a moment of maximum fragmentation and maximum consequence. Its role is not ceremonial — it is structurally necessary. But only if it operates with clarity about what multilateral forums can and cannot do. First, the Dialogue must function as the world's primary norm-convergence engine. No single nation, bloc, or company can set legitimate global AI standards unilaterally. The EU has regulatory reach, the US has innovation dominance, and China has deployment scale — but none has universal legitimacy. The Dialogue is the only forum where all three pressures meet. Its most urgent contribution is translating competing national frameworks into interoperable governance standards that reduce fragmentation without demanding uniformity. Second, it must serve as an early warning system for governance failures. The Independent International Scientific Panel on AI provides the evidence base. The Dialogue must have the institutional courage to act on that evidence — naming risks, identifying gaps, and holding member states accountable even when the most powerful actors resist. A forum that only validates consensus is not governance — it is theatre. Third, the Dialogue must democratise governance participation. Currently, AI standard-setting happens in rooms dominated by a handful of technologically advanced nations and large corporations. The Dialogue's unique value is its universal membership. Developing nations must not merely observe — they must shape norms that reflect their economic realities, cultural contexts, and development priorities. Capacity-building and governance participation must be treated as inseparable. Fourth, and most critically, the Dialogue must remain open to re-engagement. The absence of the United States is a structural weakness, not a permanent condition. The Dialogue should design its frameworks to be compelling enough that non-participating powers eventually find exclusion more costly than engagement. The Dialogue's role is not to govern AI. It is to make governance possible.

The AI Dialogue will succeed or fail based on one variable — whether it generates decisions or merely generates documents. Format and structure are not administrative details. They are the difference between governance and performance. Governments must come as negotiators, not narrators. National delegations should arrive with defined positions, red lines, and genuine flexibility — not prepared statements recycled from domestic policy speeches. The Dialogue should require member states to submit specific governance commitments, not aspirational language, as a condition of meaningful participation. The private sector must be structurally integrated, not periodically consulted. The companies building and deploying AI — disproportionately American, Chinese, and European — possess technical knowledge that no government delegation fully replicates. A formal private sector advisory track with defined input mechanisms, transparent disclosure requirements, and conflict of interest safeguards must be built into the Dialogue's architecture from the start. Consultation after decisions are made is not participation. Civil society and affected communities must have guaranteed floor time, not margin notes. Indigenous communities, journalists, healthcare workers, educators — the people experiencing AI's consequences most directly — must have structured representation. Not token panels. Binding consultation mechanisms with documented response requirements from member states. Academia and the Independent Scientific Panel should function as the Dialogue's immune system — identifying when political consensus diverges from technical reality and saying so publicly. Their independence must be protected structurally, not just rhetorically. On format, the Dialogue must abandon the traditional UN conference model of plenary speeches followed by negotiated communiqués. It needs working groups with real mandates, intersessional accountability mechanisms, and published progress tracking between Geneva and New York sessions. The structure must match the urgency. Annual meetings producing non-binding summaries will not govern the fastest-moving technology in human history.

Global AI governance discussions currently reflect a profound representation deficit. The communities with the least power to shape AI development are frequently those bearing its most significant consequences. Closing this gap is not a diversity exercise — it is a governance necessity. Unrepresented perspectives produce incomplete frameworks that fail in deployment. The Global South is the most structurally underrepresented bloc. Nations across Africa, Southeast Asia, Latin America, and the Pacific are rapidly becoming AI deployment destinations without being governance architects. They receive systems built on datasets, values, and economic assumptions that do not reflect their realities. The Dialogue must establish dedicated Global South working groups with genuine agenda-setting authority — not observer status dressed as participation. Targeted funding mechanisms must cover not just travel but sustained policy capacity development. Indigenous communities represent a near-total absence. AI systems are already making decisions affecting land rights, cultural preservation, linguistic survival, and resource allocation in indigenous territories worldwide. These communities hold governance knowledge — particularly around collective rights and intergenerational accountability — that formal state-based frameworks consistently fail to incorporate. Formal indigenous consultation protocols must be mandated, not optional. Women and gender-diverse communities remain underrepresented both as governance participants and as subjects of AI impact analysis. Facial recognition failures, algorithmic hiring bias, and content moderation disparities disproportionately affect women and marginalised genders. Gender-disaggregated impact assessments should be a baseline requirement for all Dialogue outputs. Small and medium enterprises are invisible in governance conversations dominated by large technology corporations yet face the most acute compliance burdens from fragmented regulatory frameworks. Frontline workers — healthcare professionals, educators, journalists, factory workers — are experiencing AI's operational consequences daily without any structured voice in how governance frameworks are designed. Inclusion without decision-making authority is decoration. The Dialogue must build representation into its power structure, not its press releases.

The most valuable lesson from existing AI governance efforts is simple - frameworks that connect principles to enforcement mechanisms work. Frameworks that stop at principles do not. The Dialogue should learn from both categories with equal seriousness. The EU AI Act represents the most comprehensive attempt to translate governance principles into binding obligations. Its risk-tiered approach - distinguishing unacceptable, high, limited, and minimal risk applications - provides a replicable architectural model. Its extraterritorial reach demonstrates that determined regulatory frameworks can shape global industry behavior even without universal adoption. Its limitation - compliance burden disproportionately affecting smaller economies and enterprises - is an equally important lesson. The NIST AI Risk Management Framework in the United States provides a voluntary but technically rigorous methodology for identifying, measuring, and managing AI risks across the full development lifecycle. Its adoption by US federal agencies and major corporations demonstrates that voluntary frameworks can achieve significant penetration when designed with genuine operational utility.