Indep

A successful first Global Dialogue on AI Governance would move participants toward common ground on shared standards – enough mutual understanding to enable genuine cooperation. It would be meaningfully plural and inclusive, ensuring that voices from the Global South, civil society, and communities most affected by AI have real agenda-setting power, not token seats. It would produce accountable structures with named responsibilities rather than aspirational principles. And it would foster healthy, substantive debate – particularly around under-explored alternatives like decentralised and non-extractive technologies – so that the dominant paradigms of AI development are genuinely interrogated, not simply ratified on a global stage.

• AI capacity-building
• Interoperability of governance approaches
• Protection and promotion of human rights
• Transparency, accountability, and human oversight

These selections focus on a structural gap in current digital systems: while identities and permissions are increasingly well defined, the conditions under which an entity is legitimately able to act remain largely implicit. Strengthening capacity-building ensures institutions can understand and implement models where authority and consent are explicit rather than assumed. Interoperability is essential because different legal, institutional, and community frameworks express authority in incompatible ways, limiting coordination and trust across systems. Human rights protections depend not only on principles, but on whether systems can correctly attribute and constrain action in practice. Finally, transparency and accountability require more than audit trails; they depend on making the basis of action - capacity, authority, and consent - visible and verifiable at the point of execution. Together, these priorities address the foundations of lawful and responsible digital action.

A critical emerging issue not fully captured is the lack of a clear, interoperable model for capacity and standing of the actor (or ID holder). Most current frameworks focus on identifying entities or verifying attributes, but they do not adequately express whether an entity is capable of acting in a given context, nor the basis on which that action is considered valid or binding. Capacity is contextual, time-bound, and often role-dependent. An individual or system may act personally, as an agent, as a trustee, or under delegated authority, each with different implications. Without explicit modelling of these distinctions, systems default to treating control of a credential or account as sufficient for action. This creates a structural risk where actions are technically valid but lack legitimate standing. Standing is equally underdeveloped in digital environments. It defines the basis on which an actor's actions can be recognised, relied upon, or contested by others. In practice, standing arises from a combination of role, mandate, consent, and context, yet these elements are rarely expressed together in a verifiable form. As a result, systems struggle to determine when an action should carry legal, organisational, or social effect. This gap becomes more pronounced as AI systems and autonomous agents participate in decision-making and execution. Without a clear model of capacity and standing, it becomes difficult to distinguish between authorised and unauthorised action, or to assign responsibility when outcomes are contested. Addressing capacity and standing as first-class concepts would enable more precise, accountable, and interoperable digital systems. It would also strengthen human oversight by ensuring that actions are not only attributable, but grounded in a clearly defined basis for authority and legitimacy.

In Aotearoa New Zealand and across the wider Asia–Pacific civic and governance sector, the core gap is not access to digital systems, but the absence of clear models for capacity and standing within them. Government platforms, NGOs, and emerging DAO-style initiatives can identify actors and assign permissions, yet they rarely define in what capacity those actors are operating or by what authority their actions bind others. This creates ambiguity in decision-making, delegation, and accountability, particularly in cross-agency and cross-jurisdictional work. A significant challenge is the rapid introduction of AI systems and agents into operational environments without corresponding governance primitives. AI tools are increasingly used to draft communications, make recommendations, and in some cases trigger actions, yet they operate without a clearly defined mandate. An agent may act on behalf of an organisation or individual, but the scope, limits, and revocability of that authority are not formally expressed. This creates real risk: actions can be executed efficiently and at scale, while remaining misaligned with the intent or authority of the human actors they represent. In the human rights and civil society space, this gap affects the ability to ensure that participation, consent, and representation are legitimate. Without explicit capacity and standing, it becomes difficult to verify whether decisions reflect genuine mandate or simply system-level permissions. At the same time, there is a strong opportunity. The region is actively exploring digital public infrastructure, decentralised governance models, and cross-border collaboration. Introducing clearer, interoperable ways to express capacity, authority, and consent would strengthen trust, reduce governance friction, and enable safer adoption of AI and automation. It would also support more inclusive participation by making the basis of action transparent and contestable, rather than assumed or embedded within opaque systems.

The AI Dialogue can play a critical role by aligning international actors around shared governance primitives rather than competing frameworks. By focusing on how capacity, authority, and consent are expressed and verified across systems, it can enable interoperability between legal, institutional, and technical approaches. This creates a common language for cooperation without requiring uniform regulation. It can also surface risks such as AI agents acting without clear mandate, and support the development of standards that ensure actions are attributable, authorised, and revocable across jurisdictions.

AI Dialogue shuild on what already has traction, but connect the pieces that don't currently speak to each other. That means linking high-level frameworks like the OECD AI Principles, UNESCO's AI ethics work, and the G7 Hiroshima Process with technical ecosystems such as W3C DIDs and Verifiable Credentials, Trust over IP, and access-control models like Zanzibar or Cedar. At the same time, there's real experimentation happening in DAOs, digital public infrastructure, and civil society platforms that should also be considered. Right now, these efforts sit in layers that don't quite meet. Principles talk about responsibility and rights, while technical systems manage identity and permissions, but neither clearly defines who is actually authorised to act in a given moment. That gap becomes dangerous as AI agents start operating across systems without a clear mandate. The AI Dialogue can add value by stitching these layers together. It can help develop a shared language for capacity, authority, and consent that works across legal, technical, and institutional contexts. Not another framework, but connective tissue. If it gets that right, it reduces fragmentation, makes cooperation practical, and ensures that AI-driven actions are not just possible, but legitimate and accountable.

• Different stakeholders should contribute through structured, role-specific inputs: governments framing policy constraints and interoperability needs
• technical communities defining implementable standards
• civil society grounding legitimacy, rights, and real-world risks
• and industry testing operational viability. The Dialogue should be organised as a layered process: (1) problem definition grounded in concrete use cases, (2) co-development of shared governance primitives such as capacity, authority, and consent, and (3) pilot implementations across jurisdictions. A modular, working-group format with short feedback loops and cross-sector synthesis sessions would ensure outputs remain practical, testable, and internationally relevant.

Different stakeholders should come in as co-authors of the system. Governments articulate where mandate and accountability must hold. Technical communities translate those constraints into working primitives. Civil society keeps the question of legitimacy alive, grounding it in lived reality. Industry stress-tests everything against scale and speed. People power defines the outputs and moral constraints. The Dialogue should operate like a studio. Begin with real scenarios where authority becomes unclear. Develop shared concepts of capacity, mandate, and consent from those situations. Move into live pilots across jurisdictions and sectors, with tight feedback loops and open synthesis sessions. Keep it iterative and grounded so each cycle produces something usable, testable, and capable of travelling across contexts.

Move beyond panels and position the Dialogue as a live build environment. Use scenario labs where mixed groups work through real cases – an AI agent issuing payments, moderating content, or acting for an organisation – and are forced to define capacity, mandate, and consent in practice. Pair this with team sessions to pressure-test assumptions and expose where authority breaks down. Introduce protocol studios: short, focused sprints where legal, technical, and policy participants co-draft simple, testable primitives that can plug into existing systems. Follow with pilot corridors, where a small number of cross-border use cases are trialled in real conditions with rapid feedback cycles. Layer in open witness sessions where civil society and affected communities respond to proposed models, keeping legitimacy grounded. Finally, maintain a living repository of outputs – not reports, but reusable schemas, patterns, and tested approaches. Keep it tight, iterative, and slightly loose - rough consensus and plurality in interpretation.

Several existing approaches provide strong foundations for effective AI governance, particularly where they move from principle into practice. At the policy level, frameworks such as the OECD AI Principles and UNESCO's Recommendation on the Ethics of AI establish shared expectations around accountability, transparency, and human oversight. These create a common baseline across jurisdictions, even as implementation continues to evolve. On the technical side, W3C Decentralized Identifiers (DIDs) and Verifiable Credentials (VCs) enable portable, user-controlled identity and selective disclosure. Trust over IP extends this by linking technical standards with governance frameworks, supporting interoperability across systems. Access-control models such as Google Zanzibar and OpenFGA demonstrate how relationships and permissions can be expressed with precision at scale. Emerging infrastructure such as mesh networks and decentralised physical infrastructure networks (DePIN) further expand this landscape, enabling more resilient, community-operated systems that reduce reliance on centralised platforms. These models introduce new governance considerations around participation, resource allocation, and local authority. In parallel, decentralised governance experiments, including DAO-based models and digital public infrastructure initiatives, are testing new forms of coordination and participation in real-world settings. Across these developments, governance strengthens when identity, authority, and consent are made explicit, testable, and adaptable. This gap is the focus of ongoing work developing capacity-aware identity primitives that make authority, consent, and standing explicit, auditable, and portable across digital environments.