SMU Centre for Digital Law

Success for the first Global Dialogue should be measured by agreement on practical, interoperable mechanisms. Drawing from the DPI Safeguards initiative (Global Digital Compact), five concrete outcomes would signal success: Trust as a design mandate – Adoption of a framework requiring independent audits and legislative red lines (e.g., prohibiting unauthorized surveillance) as non‑negotiable conditions for high‑risk public AI systems. Operational justice benchmarks – Move beyond normative principles to agree on justice‑readiness criteria covering procedural, distributive, and epistemic dimensions, with measurable thresholds for public‑sector AI deployment. Federated architectures as a default – Endorsement of decentralized data exchange layers to prevent vendor lock‑in and reduce systemic risk, mirroring the federated models proven in DPI. "Phygital" as a legal requirement – A commitment that AI‑driven public services maintain mandatory analog alternatives, ensuring that digital inclusion is not a precondition for accessing essential services. A Public Sector AI Use Register – Launch of a transparent, evolving register for high‑stakes AI applications (healthcare, social protection, trade), creating accountability that grows in parallel with technical deployment. These outcomes would turn the Global Dialogue into a catalyst for governance‑by‑design—delivering interoperability, building trust, and ensuring that AI governance keeps pace with deployment.

• Safe, secure and trustworthy AI
• Interoperability of governance approaches
• Transparency, accountability, and human oversight
• Open-source software, open data and open AI models

These four priorities are mutually reinforcing and reflect lessons from the DPI Safeguards initiative. Interoperability is essential to avoid fragmentation; the Global Dialogue can deliver practical value by promoting federated architectures and common standards that allow systems to work across borders without centralizing control. Transparency, accountability, and human oversight translate into concrete mechanisms: independent audits, legislative red lines, and mandatory Public Sector AI Use Registers. These ensure that oversight is not an afterthought but a design requirement. Protection and promotion of human rights grounds the entire framework-moving from high-level principles to operational justice benchmarks (procedural, distributive, epistemic) that can be measured and enforced. Safe, secure and trustworthy AI ties them together: safety and trust are built through the other three, not achieved by technical measures alone. Together, these priorities enable a "governance-by-design" approach that aligns with A/RES/79/325's call for practical cooperation.

1. Governance-by-design as a unifying framework The listed themes are often treated as separate tracks. Success requires embedding them into a single operational logic: governance mechanisms (transparency, interoperability, human rights) must be designed into AI systems from the start, not layered on after deployment. This "governance-by-design" approach should be an explicit cross-cutting principle across all thematic areas. 2. The "phygital" imperative AI-driven public services risk excluding those without connectivity or digital literacy. A cross-cutting safeguard is the legal requirement to maintain physical or analog alternatives. This issue cuts across capacity-building, human rights, and interoperability but is not fully captured by any single listed theme. Embedding it as a binding design principle would ensure that AI governance advances inclusion rather than deepening divides.

From the perspective of digital public infrastructure (DPI) deployments, governance gaps in AI are already creating measurable risks and opportunities. Challenges Interoperability gaps result in fragmented AI tools that are not portable across jurisdictions. Governments face vendor lock‑in, and smaller economies risk becoming unregulated testing grounds. Transparency and accountability deficits are most visible in high‑stakes uses—social benefit algorithms, predictive analytics, and automated health decisions. In several countries, no public register exists to track which AI systems are deployed or under what oversight. Human rights protections have not kept pace. AI tools layered onto DPI data systems often operate without clear legislative red lines on surveillance, consent, or redress, disproportionately affecting vulnerable groups. Safety and trust suffer as a direct result. Adoption stalls when citizens cannot verify how automated decisions affecting their lives are made. Opportunities DPI's federated architecture offers a proven model for interoperable AI governance: modular, decentralized data layers that prevent monopolies and enable shared accountability across borders. The DPI Safeguards framework, endorsed under the Global Digital Compact, provides a ready template to embed transparency, human rights, and safety into AI systems from the outset. There is growing demand from member states for operational tools—mandatory public registers, independent audits, and "phygital" access requirements—rather than high‑level principles alone. The Global Dialogue can seize this demand by converting DPI safeguards into concrete, replicable AI governance mechanisms.

The AI Dialogue can advance international cooperation by shifting the focus from norm‑setting to operational convergence. Member states already agree on high‑level principles; the gap lies in interoperable mechanisms that translate those principles into accountable practice. Three roles stand out: 1. A platform for "governance‑by‑design" prototypes The Dialogue can curate and endorse concrete governance tools—mandatory public registers for high‑risk AI, independent audit requirements, and "phygital" access mandates—that can be adopted across jurisdictions. Drawing on proven models like the DPI Safeguards framework, it can offer member states ready‑made, adaptable templates rather than abstract guidance. 2. A catalyst for technical interoperability International cooperation often stalls on incompatible standards. The Dialogue can promote federated architectures and common data‑exchange protocols, ensuring that AI governance approaches remain interoperable without requiring centralized control. This reduces fragmentation and lowers barriers for developing economies. 3. A mechanism for accountability alignment Cooperation requires mutual trust. By advancing commitments to legislative red lines (e.g., prohibitions on unauthorized surveillance) and justice‑readiness benchmarks, the Dialogue can help align accountability expectations across regions—making cross‑border AI deployments and collaborations more feasible. Critically, the Dialogue should not operate in isolation. It must maintain strong links with the Independent International Scientific Panel on AI and the DPI Safeguards track under the Global Digital Compact. By connecting scientific evidence, operational safeguards, and diplomatic cooperation, the Dialogue can deliver the practical interoperability that member states increasingly demand.

The AI Dialogue should build upon and connect with: Global Digital Compact & DPI Safeguards initiative – Provides operational safeguards for digital infrastructure; offers proven templates (public registers, federated architectures, phygital mandates) that can be adapted for AI governance. Independent International Scientific Panel on AI – Established under the same resolution (A/RES/79/325); the Dialogue can translate its scientific findings into actionable policy mechanisms. UNESCO Recommendation on AI Ethics – Delivers normative framework; the Dialogue can add operational benchmarks (e.g., justice-readiness criteria) that make ethics enforceable. OECD AI Principles & GPAI – Offer multi-stakeholder collaboration and interoperability guidance; the Dialogue can embed these into a UN-wide interoperability agenda. EU AI Act and regional frameworks – Provide regulatory models; the Dialogue can facilitate cross-regional alignment and help developing economies adapt similar safeguards. ITU's AI for Good – Showcases practical AI applications; the Dialogue can ensure that such initiatives operate within agreed accountability and human rights guardrails.

Member states – Provide political endorsement for interoperable mechanisms (e.g., public registers, audit requirements) and pilot them domestically, demonstrating feasibility. Civil society & academia – Contribute independent monitoring, justice-readiness benchmarks, and real‑world impact assessments to ensure accountability and equity. Private sector & technical community – Offer expertise on federated architectures, open standards, and security‑by‑design; commit to transparency through voluntary adoption of public use registers. DPI practitioners & digital public infrastructure implementers – Share operational safeguards (phygital mandates, legislative red lines) as templates that can be adapted for AI governance. International organizations & regional bodies – Align existing frameworks (UNESCO, OECD, ITU) into a coherent interoperability agenda, avoiding duplication. Format and structure recommendations Thematic working groups – Establish focused groups corresponding to the four priorities (interoperability, transparency/oversight, human rights, safety/trust) to develop concrete deliverables, not just discussions. "Governance‑by‑design" labs – Structured, multi‑stakeholder sessions where member states, technologists, and civil society co‑design operational tools (e.g., public register templates, audit protocols) that can be adapted across jurisdictions. Linked dialogue cycles – Synchronize with the Independent International Scientific Panel on AI: the Panel provides evidence and risk assessments; the Dialogue translates them into policy mechanisms and tracks implementation. Public commitment registry – A transparent, evolving record where stakeholders voluntarily register their AI governance commitments (e.g., adopting a public register, mandating audits) to build accountability and track progress. Hybrid and regional anchors – Use hybrid meetings to broaden participation, with regional hubs to ensure diverse geographic input and reflect local contexts in global standards.

Underrepresented voices Digital public infrastructure (DPI) practitioners – Those implementing large‑scale identity, payments, and data exchange systems in developing economies possess frontline insights on inclusion, exclusion risks, and operational safeguards, yet are often absent from global policy forums. Users of AI‑mediated public services – Citizens—particularly marginalized groups, persons with disabilities, and those in low‑connectivity settings—experience the real‑world impacts of AI governance gaps but rarely inform the design of oversight mechanisms. Grassroots civil society and community‑based organizations – Local groups that mediate trust, redress, and digital literacy operate outside formal UN and industry networks. Small island developing states (SIDS) and least developed countries (LDCs) – Their governance needs (e.g., avoiding vendor lock‑in, ensuring analog alternatives) differ from larger economies, yet they are often underrepresented in agenda‑setting. Technical implementers from the Global South – Engineers, architects, and privacy experts building federated, open‑source AI‑ready systems offer practical "governance‑by‑design" experience that complements high‑level policy perspectives. How to include them Targeted participation support – Provide dedicated funding, translation, and capacity support to enable DPI practitioners, grassroots organizations, and LDC/SIDS representatives to attend and contribute meaningfully. Regional consultation hubs – Host structured consultations in Africa, Asia‑Pacific, Latin America, and the Caribbean, feeding outcomes into the central Dialogue rather than relying on self‑selection. User‑centered mechanisms – Mandate participatory design sessions where affected communities (e.g., social benefit recipients) validate justice‑readiness benchmarks and oversight tools. Open, asynchronous inputs – Supplement high‑level meetings with accessible online platforms that allow written, audio, or video submissions from practitioners and citizens, ensuring time‑zone and connectivity barriers do not limit engagement. Link to DPI Safeguards networks – Leverage existing DPI implementation networks to channel operational expertise directly into the Dialogue's working groups.

To move beyond traditional panels and statements, the AI Dialogue should adopt formats designed for co‑creation, experiential learning, and sustained collaboration. 1. Governance‑by‑Design Labs Structured, facilitated workshops where member states, technologists, civil society, and affected communities co‑design operational tools—such as public register templates, audit protocols, or justice‑readiness benchmarks—in real time. Labs produce tangible outputs rather than merely exchanging views. 2. Justice‑Readiness Simulations Scenario‑based exercises where participants role‑play high‑stakes AI deployments (e.g., automated social benefit allocation) and pressure‑test governance mechanisms. These simulations expose gaps, build shared understanding across stakeholders, and generate practical recommendations. 3. Peer‑Learning Cohorts Small, sustained groups of member states and practitioners working together over months to implement a specific governance mechanism (e.g., a public sector AI register), with facilitated checkpoints to share lessons. Cohorts bridge the Dialogue's periodic meetings with ongoing implementation. 4. Digital Sandboxes Interactive, secure environments where participants experiment with federated data architectures, open‑source audit tools, or algorithmic impact assessment frameworks. Sandboxes demystify technical concepts and enable hands‑on learning for policymakers. 5. Community‑Based Feedback Loops Structured channels that feed input from underrepresented groups—DPI practitioners, grassroots organizations, and service users—directly into working groups, with dedicated sessions to validate findings and adjust recommendations before finalization. 6. Open Commitments Tracker A public, interactive platform where stakeholders voluntarily register governance commitments (e.g., adopting a public register, mandating phygital access). The tracker fosters accountability and enables others to replicate proven approaches.

1. Public Sector AI Use Registers Modeled on DPI transparency frameworks, these registers mandate public disclosure of AI systems used in high-stakes areas (social protection, health, policing). Examples: New York City's Automated Decision Systems (ADS) Law and emerging registers in the EU AI Act provide templates. They enable audits, redress, and public accountability. 2. Legislative "Red Lines" in DPI Safeguards Several DPI-implementing countries have codified prohibitions on unauthorized surveillance, data sharing without consent, and use of digital ID for political profiling. These "red lines" offer a proven approach to embedding human rights into digital and AI systems from the outset. 3. Federated Data Exchange Architectures India's IndiaStack, Estonia's X-Road, and open-source platforms like MOSIP (Modular Open Source Identity Platform) demonstrate how federated, decentralized data layers prevent vendor lock-in, enable interoperability, and reduce systemic risk-a model directly applicable to AI governance. 4. "Phygital" Service Delivery Mandates Brazil's Digital Government Strategy and Colombia's digital inclusion laws require that essential public services remain accessible through physical channels, ensuring AI-first does not exclude analog citizens. This safeguards inclusion and human rights. 5. Independent Algorithmic Audit Frameworks Canada's Directive on Automated Decision-Making mandates algorithmic impact assessments and peer audits before deployment. The Netherlands' algorithm registry and civil society-led audits (e.g., AlgorithmAudit.org) provide operational models for transparency and oversight. 6. Open-Source AI Governance Tools Initiatives like the AI Incident Database, OECD's AI Policy Observatory, and the Global Index on Responsible AI offer open platforms for tracking risks, sharing governance practices, and enabling collective learning across jurisdictions.