The Inniss Institute for Digital Policy and Intellectual Property

Success for the first Global Dialogue on AI Governance should be measured not by the ambition of its declarations but by the precision of its mandates. The international community has produced sufficient normative consensus on AI principles. What it has not produced is the architectural infrastructure to make those principles effective in the jurisdictions that need them most. On that basis, three outcomes would constitute genuine success. First, the Dialogue should produce formal recognition that the governance deficit in the Global South is a policy architecture failure, not a technology or finance deficit. This distinction matters because it redirects remedies: the solution is not more national AI strategies or additional connectivity funding, but targeted investment in the enabling instruments — statutory frameworks, regulatory capacity, judicial competency — that translate normative commitments into enforceable outcomes. Second, the Dialogue should mandate the development of a UN Sovereign Data Rights Standard establishing that data generated within a jurisdiction constitutes a sovereign asset of that jurisdiction. The absence of this principle from current international instruments is the juridical condition that enables the systematic extraction of data from developing economies without compensation, consent, or legal recourse. Naming and closing that gap would represent a structurally significant advance. Third, the Dialogue should endorse a process for developing a Sovereign Consent Protocol modelled on the Nagoya Protocol on Access and Benefit-Sharing, establishing that access to nationally generated data for AI training purposes is conditional upon prior sovereign consent and a binding benefit-sharing agreement. A Dialogue that produces these three mandates — even in preliminary form — will have moved the international community from aspiration to architecture. That is the standard against which success should be judged.

• Social, economic, ethical, cultural, linguistic and technical implications of AI
• AI capacity-building
• Protection and promotion of human rights
• Interoperability of governance approaches

Our four selections reflect a single coherent argument: that equitable AI governance requires structural intervention at the level of juridical architecture, not only technical standards or ethical principles. Social, economic, ethical, cultural, linguistic and technical implications of AI is our primary priority because it encompasses the condition we have termed Inniss Data Nullius - the juridical absence of domestic frameworks recognising locally generated data as a sovereign, rights-bearing asset. The systematic extraction of cultural, linguistic, health, and economic data from small island developing states and Global South jurisdictions without compensation, consent, or legal recourse is the most consequential implication of AI development currently unaddressed by international instruments. It cannot be remedied by ethics frameworks alone; it requires legal architecture. Protection and promotion of human rights follows directly. The right to cultural integrity, the right of peoples to benefit from their own resources, and the right to self-determination are implicated when a jurisdiction's data - its language, its health patterns, its artistic heritage - is processed into commercially valuable AI systems with no legal obligation to the originating community. These are not incidental harms; they are structural features of the current AI development model. AI capacity-building is selected because our research demonstrates that the prevailing model of capacity support - funding national AI strategies and data protection legislation - systematically fails to address the enabling architecture that makes governance effective. Regulators without enforcement tools, courts without judicial guidance, and legislators without secondary instruments produce ghost laws, not governance. Reorienting capacity-building toward this enabling infrastructure is urgent. Interoperability of governance approaches is essential because the Sovereign Consent Protocol we propose - conditioning AI training data access on prior sovereign consent and benefit-sharing - can only function within an interoperable international framework. Fragmented approaches entrench the advantage of actors already operating at global scale.

Yes. The listed thematic areas, taken together, address what AI governance should achieve. What they do not adequately address is why governance commitments systematically fail to produce those outcomes in the jurisdictions that need them most. That gap is itself a cross-cutting issue requiring explicit treatment. We term it the Execution Gap: the structural disconnect between the adoption of international AI governance standards and their functional, locally-governed enforcement. It is cross-cutting because it operates beneath every thematic area listed. A jurisdiction may adopt safe and trustworthy AI principles but lack the regulatory infrastructure to enforce them. It may enact human rights protections for AI-affected communities but have no judicial capacity to adjudicate claims. It may commit to transparency and accountability but have no statutory instruments defining what those obligations require of actors operating within its territory. The Execution Gap is the reason normative progress at the international level does not translate into governance outcomes at the jurisdictional level. The second cross-cutting issue not captured by the listed themes is jurisdictional data rights - the legal question of whether data generated within a state constitutes a sovereign asset of that state. This is foundational to every other governance objective. Capacity-building cannot produce economic sovereignty if the data economy's primary inputs remain legally ownerless under domestic law. Human rights protections cannot reach AI harms if the data generating those harms was extracted without any legal obligation to the originating jurisdiction. Interoperability cannot produce equity if the parties to governance agreements do not hold equivalent legal standing over their own data resources. Both issues - the Execution Gap and the absence of jurisdictional data rights - are preconditions for the listed thematic areas to function as intended. We respectfully submit that the Dialogue will be most effective if it addresses these architectural preconditions explicitly, rather than treating them as assumed.

The governance gaps identified in our selected thematic areas manifest with particular acuity in Caribbean and small island developing state jurisdictions, where limited administrative capacity, small domestic markets, and deep integration into global digital infrastructure compound the structural vulnerabilities that larger economies can partially offset. The most significant challenge is the condition we have termed Inniss Data Nullius: the absence of domestic legal frameworks recognising locally generated data as a sovereign, rights-bearing asset. Caribbean populations generate substantial volumes of economically valuable data — health and epidemiological records, linguistic and cultural production, agricultural and environmental datasets, tourism behavioural data — that are routinely extracted, processed, and commercialised by foreign-domiciled AI developers without consent, compensation, or legal obligation to the originating jurisdiction. The derived intellectual property is registered offshore. The commercial value does not return. Under current international instruments, there is no legal mechanism to compel otherwise. This is compounded by the Execution Gap. Caribbean jurisdictions have participated actively in regional and international digital governance forums and have enacted data protection and AI-related legislation. But the enabling architecture — specialist regulatory capacity, secondary instruments, judicial training, enforcement mechanisms — has not kept pace. The result is a regional landscape of ghost laws: formally adequate frameworks that produce no substantive regulatory outcomes because the institutional infrastructure to implement them does not exist. The opportunity is equally significant. Caribbean jurisdictions are not starting from zero. Regional institutions — including CARICOM and the Caribbean Court of Justice — provide an existing architecture for collective action. The region's experience of twenty years of implementation failure constitutes a detailed evidence base for what does not work and what enabling architecture actually requires. That experience, properly theorized and documented, positions the Caribbean as a credible contributor to the design of international governance instruments rather than merely a recipient of them. This submission is offered in that spirit.

The AI Dialogue occupies a position in the international governance landscape that no other forum currently fills: it is the only space where the full membership of the United Nations can engage collectively on AI governance as a matter of shared sovereign concern rather than technical standard-setting or bilateral negotiation. That position gives it three distinct roles that no other body can play. The first is legitimation. Governance principles that emerge from an inclusive intergovernmental dialogue carry a form of political authority that OECD guidelines, G7 communiqués, and industry-led frameworks cannot replicate. For the Global South, this distinction is not procedural — it is substantive. The current international AI governance landscape is architecturally skewed toward the preferences and institutional capacities of high-income jurisdictions. A UN-anchored dialogue is the primary mechanism available to rebalance that architecture toward universal participation and equitable outcomes. The second is standard-setting with jurisdictional reach. The Dialogue is positioned to mandate the development of instruments — a UN Sovereign Data Rights Standard, a Sovereign Consent Protocol — that would establish binding expectations for how AI developers interact with nationally generated data across jurisdictions. No technical body or voluntary framework can produce instruments with equivalent reach or enforceability. The third is architectural coordination. The most consequential governance failures we have identified are not failures of principle but failures of implementation infrastructure. The Dialogue can play a coordinating role that bilateral technical assistance cannot: aligning the work of UNCTAD, WIPO, ITU, UNDP, and regional bodies toward a shared enabling architecture rather than parallel and frequently duplicative national capacity-building programmes. The Dialogue's unique value is not consensus on principles — that consensus largely exists. It is the conversion of consensus into mandates, and mandates into architecture. That conversion is where international cooperation on AI governance will either succeed or stall.

Several existing instruments and initiatives provide foundations the Dialogue should explicitly build upon rather than duplicate. The Nagoya Protocol on Access and Benefit-Sharing under the Convention on Biological Diversity is the most directly relevant precedent. It established that access to genetic resources for commercial purposes requires prior informed consent and equitable benefit-sharing with the country of origin. The architecture of a Sovereign Consent Protocol for AI training data is structurally analogous. The Dialogue should commission a formal comparative legal analysis of the Nagoya Protocol as a design template, engaging the CBD Secretariat as an institutional partner. UNCTAD's work on digital economy and data governance, particularly its Digital Economy Reports, has produced the most rigorous available analysis of data flow imbalances between the Global North and Global South. The Dialogue should formally incorporate UNCTAD's evidence base into its deliberations and connect its standard-setting work to UNCTAD's ongoing policy development on cross-border data flows. WIPO's development agenda addresses the relationship between intellectual property systems and development equity. The condition of Inniss Data Nullius — in which AI-derived intellectual property is registered offshore on the basis of data extracted from developing jurisdictions — is directly within WIPO's mandate. The Dialogue should engage WIPO to examine whether existing IP instruments require amendment to address this dynamic. Regional governance architectures — including CARICOM, the African Union's data governance frameworks, and ASEAN's digital economy agreements — represent implementation infrastructure that global instruments must be designed to connect with rather than override. The Dialogue's added value here is coordination: ensuring that regional frameworks are mutually reinforcing and that smaller regional bodies have pathways to participate in global standard-setting on equitable terms. The Dialogue's unique added value across all of these is jurisdictional authority — the capacity to convert existing analytical work and regional experimentation into instruments that carry the weight of universal membership.

The structure of the Dialogue will determine whose knowledge shapes its outcomes. Format is not a procedural question — it is a power question. Our recommendations address both. On stakeholder contribution, the Dialogue should distinguish between three categories of participant whose contributions serve different functions and should be structured accordingly. Member State delegations carry sovereign authority and are the only actors who can convert dialogue outcomes into binding instruments. Their engagement should be focused on the three proposals advanced in this submission: recognition of the Execution Gap, a Sovereign Data Rights Standard, and a Sovereign Consent Protocol. Procedurally, delegations from SIDS and low-income jurisdictions require dedicated speaking time and pre-session briefing support to participate on substantively equal terms with larger delegations. Specialised research institutions and policy architects — including institutes working on digital governance in SIDS and Global South contexts — hold the implementation knowledge that diplomats cannot be expected to carry. Their contribution is most valuable not in plenary statements but in technical working groups tasked with translating political mandates into draft instrument language. The Dialogue should create formal mechanisms for this contribution, distinct from general civil society participation. Civil society and affected communities hold experiential knowledge of how AI governance failures manifest in practice — the health data extracted without consent, the cultural heritage processed without attribution, the agricultural communities sold back their own environmental data as proprietary products. Their testimony should inform the evidentiary base of the Dialogue's findings, not merely its peripheral consultations. On format, the Dialogue should dedicate at least one structured session specifically to implementation architecture — not principles, but the enabling instruments required to make principles effective. This session should produce draft language, not recommendations to produce draft language. The distance between those two outcomes is precisely the Execution Gap the Dialogue exists to close.

The underrepresentation in global AI governance discussions is not accidental. It is structural, and it mirrors the extractive dynamic that the Inniss Data Nullius Framework is designed to address. The communities most affected by AI governance failures are systematically least present in the forums designing the remedies. Small island developing states are the most acutely underrepresented category. SIDS jurisdictions generate data at scale — health, environmental, cultural, linguistic, agricultural — that is incorporated into globally deployed AI systems. Yet SIDS delegations are routinely absent from technical working groups, underrepresented on expert panels, and structurally disadvantaged in negotiating processes that reward large, well-resourced delegations. Inclusion requires dedicated SIDS caucus time, funded participation, and explicit reserved seats on expert bodies — not as a courtesy but as a governance necessity, since SIDS experience constitutes irreplaceable evidence about implementation failure at the jurisdictional level. Indigenous and traditional knowledge communities whose cultural production, linguistic heritage, and ecological knowledge are being processed as AI training data have no formal standing in current governance frameworks. Their knowledge systems are simultaneously the raw material of generative AI development and entirely absent from its governance. Inclusion requires the creation of formal participation mechanisms modelled on the engagement frameworks developed under the Convention on Biological Diversity, which has more successfully incorporated indigenous knowledge holders into international deliberation than any AI governance forum to date. Practitioners of legal and policy architecture in the Global South — the lawyers, regulators, judges, and policy specialists who understand what implementation actually requires in low-capacity jurisdictions — are consistently displaced in global forums by technology companies and academic institutions from high-income countries. Their knowledge is the most operationally relevant available for closing the Execution Gap. Inclusion requires that technical working groups be explicitly composed to represent implementation experience, not only technical expertise or institutional scale. Representation is not a values question here. It is an evidence question. The Dialogue will produce better instruments if the people who understand implementation failure from the inside are in the room where instruments are designed.

The most consequential innovation the Dialogue could adopt is deceptively simple: replacing recommendation sessions with drafting sessions. The standard UN dialogue format produces communiqués that call upon expert groups to develop instruments at a future date. That format is precisely the mechanism through which normative consensus accumulates without producing governance outcomes — the Execution Gap operating at the institutional level. Sessions structured around the production of actual draft language, with legal drafters in the room alongside diplomats, would compress the distance between political mandate and actionable instrument by months or years. Jurisdiction stress-testing would be a high-value format innovation. Rather than abstract panel discussions on AI principles, structured sessions would present a specific governance scenario — a foreign AI developer extracting health data from a SIDS jurisdiction, a generative AI system trained on indigenous cultural heritage without consent — and task mixed delegations of diplomats, legal practitioners, and affected community representatives with identifying the precise point at which existing international instruments fail and what draft language would close the gap. This format generates concrete outputs and surfaces implementation knowledge that plenary debate does not reach. Asymmetric expert panels would correct the current imbalance in whose knowledge shapes dialogue outcomes. Standard panel formats aggregate similar voices — typically technologists and policy academics from high-income jurisdictions. Panels structured to place a SIDS regulatory practitioner, an indigenous knowledge holder, and a legal architect from a low-capacity jurisdiction alongside a major AI developer would generate the productive friction from which genuinely new governance thinking emerges. Pre-session instrument clinics — structured working sessions held in the weeks before the Dialogue in which SIDS and Global South delegations work with specialist legal drafters to translate their governance priorities into proposed text — would ensure that smaller delegations arrive as co-authors of the Dialogue's outputs rather than respondents to drafts produced by others.

The most instructive examples for the Dialogue are not the celebrated successes of high-income jurisdictions but the documented failure modes of low-capacity ones - because it is from failure analysis, not success replication, that genuinely transferable governance architecture is built. The Nagoya Protocol on Access and Benefit-Sharing remains the most structurally relevant international precedent for the governance challenge the Dialogue faces. By establishing that commercial access to biological resources requires prior informed consent and equitable benefit-sharing with the country of origin, it created a juridical architecture that directly addresses the extractive dynamic. Its limitations - incomplete ratification, enforcement gaps, and the difficulty of tracing derived products back to source materials - are precisely the implementation challenges a Sovereign Consent Protocol for AI training data must be designed to avoid from the outset rather than correct retrospectively. The Caribbean Court of Justice offers an instructive example of regional judicial architecture designed to develop endogenous jurisprudence rather than interpret imported legal frameworks. Its potential role in developing authoritative regional AI governance jurisprudence - analogous to the role the Court of Justice of the European Union played in data protection through the Schrems decisions - demonstrates that effective governance does not require replicating high-income institutional models but rather investing in existing regional institutions with the mandate and capacity to develop context-specific legal reasoning. The Inniss Data Nullius Framework (DOI: 10.5281/zenodo.19865512) offers a concrete policy architecture tool: a diagnostic methodology for identifying the specific point at which a jurisdiction's legal framework fails to recognize locally generated data as a sovereign asset, and a four-phase intervention design process for closing that gap through targeted enabling legislation rather than wholesale regulatory transplantation. The EU GDPR's adequacy decision architecture demonstrates both the power and the limitation of unilateral standard-setting: it has raised global data protection standards but has done so by exporting a framework designed for high-capacity jurisdictions to contexts where the enabling infrastructure does not exist - producing precisely the ghost law dynamic the Execution Gap framework identifies.