Good governance India foundation

A successful Global Dialogue on AI Governance would be one that resists the production of purely rhetorical consensus and instead confronts the structural conditions under which AI systems are designed, deployed, and controlled. Declarations of "trustworthy" or "ethical" AI are of limited value if they are not anchored in mechanisms that constrain power and enable independent verification. Success would therefore require, first, the articulation of governance principles in operational terms translating abstract commitments into enforceable system properties such as auditability, traceability, and reproducibility. Without this, governance remains symbolic. Second, the dialogue must address asymmetries in technical and institutional power. AI development and control are currently concentrated among a limited set of actors. A meaningful outcome would include concrete pathways for redistributing capacity through open infrastructure, shared standards, and institutional support so that governance is not effectively privatized. Third, the process should produce interoperable frameworks that allow systems to be evaluated across jurisdictions without deferring to dominant regulatory or corporate models. Absent this, fragmentation will be resolved not through coordination, but through the extension of existing power structures. A successful dialogue is therefore not one that achieves consensus, but one that establishes conditions under which claims about AI systems can be independently tested, contested, and constrained.

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

These priorities reflect the need to address AI governance as a problem of power, not merely design. Transparency, accountability, and human oversight are necessary because systems that cannot be examined cannot be governed. However, transparency must be understood as enabling independent scrutiny, not controlled disclosure. Interoperability of governance approaches is essential to prevent the consolidation of de facto global standards by dominant actors. Without coordination mechanisms, governance fragmentation will not produce pluralism but rather reinforce existing hierarchies. Open-source software, data, and models are critical as they reduce barriers to inspection and participation. In their absence, technical knowledge remains concentrated, and governance becomes dependent on claims that cannot be independently verified. Protection and promotion of human rights is included not as an abstract commitment, but as a constraint on system design and deployment. Rights-based language is often invoked, but without enforceable mechanisms it risks functioning as legitimizing rhetoric. Together, these priorities aim to shift governance from declarative norms to material conditions that enable scrutiny, participation, and constraint.

A central issue insufficiently captured is the problem of concentrated control over AI infrastructure. Governance discussions often assume neutral technical systems, whereas in practice, control over models, compute, and data is highly centralized. This concentration shapes not only deployment but also the terms under which governance itself is defined. A second issue is the gap between formal commitments and enforceability. Principles such as fairness or accountability are frequently articulated, yet lack mechanisms for independent verification. Without enforceability, governance risks functioning as post hoc justification rather than constraint. Third, the distinction between access and capacity is critical. Making systems or data available does not ensure that actors can meaningfully engage with them. Technical, institutional, and resource asymmetries limit who can participate in governance processes, even within formally open frameworks. Finally, there is a need to examine how governance frameworks may legitimate existing systems rather than transform them. If governance operates primarily by adapting to current power structures, it risks stabilizing them under the language of responsibility. Addressing these issues requires shifting focus from principles to structures-specifically, the distribution of control, the conditions of verification, and the capacity for independent oversight.

Yes. The governance gaps are already affecting India and other low- and middle-income contexts in concrete ways, particularly through large-scale digital public infrastructure such as Aadhaar, UPI, and emerging urban data platforms, as well as early-stage digital twin initiatives in cities like Delhi and Pune. The primary challenge is the expansion of systems that operate at population scale without corresponding mechanisms for independent verification. In India, digital systems are used for identity authentication, welfare distribution, and financial transactions. While these systems have demonstrated efficiency gains, their internal decision processes, data linkages, and error-handling mechanisms are not always fully accessible for external audit. Documented issues such as authentication failures in Aadhaar-linked services and exclusion errors in welfare delivery indicate that system performance cannot be assessed solely through aggregate success metrics. A second issue is the gap between formal transparency and practical auditability. Technical documentation, where available, does not necessarily enable independent verification. The capacity to audit complex systems remains limited to a small group of actors with specialized expertise, which restricts meaningful accountability. At the same time, there is a real opportunity. India's digital public infrastructure is built on relatively modular and API-driven architectures, and in some cases uses open standards. This creates a pathway for embedding verifiability into system design through stronger logging, audit trails, and reproducibility mechanisms. The urgency is structural. As digital systems extend into urban governance through digital twins and data platforms, the absence of verifiable design will scale existing accountability gaps. The issue is not whether these systems function, but whether their operation can be independently examined, tested, and contested in practice.

he AI Dialogue can play a meaningful role only if it shifts cooperation from declarative alignment to operational coordination. At present, international discussions on AI governance tend to converge at the level of principles while diverging in implementation. This creates a gap where systems are governed differently in practice despite nominal agreement. The Dialogue can address this by establishing shared technical reference points. Instead of attempting uniform regulation, it can define interoperable standards for auditability, documentation, risk classification, and system evaluation. This would allow different jurisdictions to maintain regulatory autonomy while ensuring that AI systems can be assessed across borders using comparable criteria. A second role is to reduce asymmetries in governance capacity. Many countries lack the technical and institutional resources required to evaluate complex systems. The Dialogue can facilitate cooperation through shared infrastructure, technical assistance, and open tools for testing and auditing, preventing governance from becoming dependent on a small number of actors. Third, it can function as a coordination layer between fragmented initiatives. Existing efforts often operate in isolation, leading to duplication or incompatible approaches. The Dialogue can align these efforts without subsuming them, creating a structured interface between technical, regulatory, and institutional domains. Its value, therefore, lies not in producing consensus statements, but in enabling comparability, verifiability, and coordination across different governance systems

The AI Dialogue should build upon existing initiatives that have already developed governance frameworks, technical standards, and institutional coordination mechanisms. Relevant efforts include the OECD AI Principles, which provide widely adopted normative guidelines; UNESCO Recommendation on the Ethics of Artificial Intelligence, which establishes a multilateral ethical baseline; the Global Partnership on AI, which connects research and policy communities; and the G20 AI Principles, which reflect political alignment among major economies. In addition, technical standard-setting bodies such as ISO/IEC JTC 1/SC 42 and emerging risk-based regulatory approaches, including the EU AI Act, provide concrete attempts to translate principles into enforceable requirements. However, these initiatives remain fragmented. They differ in scope, level of abstraction, and mechanisms of enforcement. The result is partial overlap without effective interoperability.

Different stakeholders can contribute meaningfully only if participation is structured beyond symbolic inclusion. The Dialogue should be organized in layered formats: technical working groups, policy coordination tracks, and public-interest forums. Each should produce concrete outputs such as standards, audit frameworks, or implementation guidelines. Participation must be role-defined, with engineers, regulators, and civil society contributing to specific problem domains rather than general discussion. Iterative cycles, draft reviews, and testable outputs are essential. Without structured contribution mechanisms, multi-stakeholder participation risks becoming procedural rather than substantive, limiting its impact on actual system design and governance.

Underrepresentation persists among actors without technical or institutional power, particularly from low- and middle-income countries, public sector implementers, and communities directly affected by large-scale systems. The issue is not only access but capacity. Inclusion requires enabling conditions: funding, technical training, and access to tools for independent evaluation. Local system operators and domain experts should be integrated into governance processes, as they encounter system failures in practice. Without addressing structural inequalities in expertise and resources, participation will remain formal, and governance outcomes will continue to reflect the perspectives of a narrow set of actors.

Effective engagement requires formats that produce verifiable outputs rather than discussion alone. Structured audit simulations, where participants evaluate real or prototype systems, can ground governance in practice. Technical-policy labs can connect system design with regulatory requirements. Red-teaming exercises can expose system vulnerabilities under controlled conditions. Iterative working sessions with defined deliverables, such as draft standards or audit protocols, should replace one-time panels. Digital collaboration platforms enabling shared documentation and version control can support continuity. Engagement should be task-oriented, evidence-driven, and repeatable, ensuring that participation contributes to measurable improvements in governance capacity.

Effective AI governance depends on enforceable constraints, not abstract principles. The EU AI Act introduces risk-based obligations and documentation requirements, shifting governance toward ex ante control. Frameworks like NIST AI Risk Management Framework and standards from ISO/IEC JTC 1/SC 42 translate principles into technical procedures, though they remain largely voluntary. Open digital infrastructures, including India's API-based systems, demonstrate scalability but raise auditability concerns. Across these examples, the limitation is consistent: transparency without verifiability and capacity does not constrain power. Effective governance requires integration of regulation, open system design, and independent audit capability.