AI Commons / Cognizant

Success for the first Global Dialogue means three things that reinforce each other. First, a shift in framing: from "what principles should govern AI" to "what mechanisms can make those principles operational." The Dialogue should mark the point where the international community acknowledged that governance is not only a coordination problem but an infrastructure problem. It requires institutional investment, technical standards, and verifiable accountability, not just agreed language. Second, a foundation for inclusivity that goes beyond representation. Success is not achieved by having diverse voices in the room. It is achieved when those voices have genuine authorship over outcomes. This means the Global South, the AI training workforce, civil society, and communities most affected by AI deployment are not consulted after the architecture is set, but they help set it. Third, a concrete commitment to governance investment as a measurable priority. The Dialogue should produce at least one mechanism - a metric, a reporting standard, a multilateral fund, or a shared capability, that treats governance infrastructure as a global public good and begins closing the gap between the resources available for AI capability development and those available for AI governance. Without this, the Dialogue risks reinforcing a pattern where principles proliferate while implementation capacity stagnates. A successful first Dialogue does not need to resolve everything. It needs to establish that the international community is serious about execution, and to leave with at least one thing that did not exist before: a shared mechanism, a binding commitment, or a governance capability that works across jurisdictions.

• Safe, secure and trustworthy AI
• Transparency, accountability, and human oversight
• Interoperability of governance approaches
• AI capacity-building

These four priorities reflect a coherent diagnosis: the core challenge in AI governance is not the absence of principles but the absence of mechanisms that enforce them reliably and at scale. Safe, secure and trustworthy AI is foundational, but "trustworthy" must be understood operationally, not aspirationally. Trust is currently the primary bottleneck for AI adoption in consequential domains. Building it requires continuous assurance, not one-time audits. Interoperability of governance approaches is urgent because the fragmentation of national and regional frameworks is creating compliance complexity that large actors can navigate and smaller actors cannot. A global floor of interoperable standards is more valuable than multiple high-quality frameworks that cannot communicate with each other. Transparency, accountability, and human oversight connects directly to the challenge of agentic AI and systems that act autonomously across multi-step processes where traditional human review does not attach in real time. Governance frameworks were largely designed for AI that produces outputs for human review. They are not yet designed for AI that acts. This gap is growing faster than it is being addressed. AI capacity-building is included because governance without institutional capacity is theater. The divide in AI is increasingly not only about who can access AI systems, but about who has the infrastructure and readiness to govern them once deployed. This includes regulatory capacity, civil society AI literacy, and "critically" the standing of the AI training workforce in governance processes.

Two cross-cutting issues deserve explicit attention. The first is the governance investment gap. There is no shared methodology for measuring how much is invested in AI governance relative to AI capability development. Without measurement, accountability is impossible. The Dialogue should advocate for a Global Governance Investment Metric analogous to R&D intensity ratios in other technology sectors. That will create a basis for comparison, accountability, and targeted support. Capital allocation is itself a form of governance, and it is currently invisible in governance discussions. The second is the status of the AI training workforce. The individuals who perform data annotation, content moderation, and AI feedback at scale, predominantly in lower-income countries, are the source of much of the knowledge that powers contemporary AI systems. They are structurally excluded from governance processes that shape how that knowledge is used. They have no formal standing equivalent to that of industry, civil society, or academia in any current multilateral AI governance forum. This is a legitimacy problem, not only an equity problem. Governance frameworks that do not account for the people whose labor underpins the systems being governed are incomplete by design. Both issues cut across all four thematic clusters. The investment gap affects capacity, safety, interoperability, and accountability simultaneously. The labor question touches human rights, inclusion, transparency, and the economic implications of AI. Neither fits neatly into a single cluster, which is precisely why they risk falling through the gaps if not named explicitly.

From the perspective of large-scale enterprise AI deployment and international governance engagement, the most significant challenge is the execution gap: the distance between what governance frameworks require and what organizations, particularly those operating across multiple jurisdictions, can actually implement in production environments. AI systems in enterprise contexts are dynamic. They adapt, drift, and interconnect with other systems in ways that periodic audits and static compliance reviews cannot track. The governance frameworks currently available define expectations but do not yet provide mechanisms to enforce them continuously once systems are deployed. This is not a criticism of the frameworks, and it reflects the genuine difficulty of governing complex adaptive systems. But it means that even organizations with strong governance intent face structural gaps between their commitments and their operational reality. At the international level, the proliferation of governance frameworks, the EU AI Act, national AI strategies, sectoral standards, voluntary commitments, are creating a fragmented landscape that is difficult to navigate consistently. The compliance burden falls disproportionately on organizations with fewer resources, and the risk of regulatory arbitrage grows as actors optimize for the least demanding jurisdiction. The opportunity is that the enterprise sector has accumulated significant practical knowledge about what governance looks like when it functions, including real-time monitoring, dynamic policy enforcement, chain-of-accountability in multi-agent systems. That knowledge has not yet been adequately translated into international standards discussions. The Global Dialogue is well-placed to facilitate that translation.

The AI Dialogue can play a role that no existing process currently fills: serving as the legitimate, universal platform where the political and the technical dimensions of AI governance are addressed together, not in parallel. Most existing governance processes operate in one of two modes. Political processes including UN General Assembly, G7, and G20 establish principles and commitments but lack the technical specificity to drive implementation. Technical processes coming from standards bodies, industry consortia, and research networks develop implementable specifications but lack the legitimacy and universality to ensure adoption across jurisdictions and sectors. The Dialogue sits at the intersection. It can do three things that would constitute genuine added value. - First, it can translate political commitments into technical requirements, identifying which principles need to become standards, and initiating the processes to develop them with universal participation. - Second, it can establish governance investment as an international priority, creating accountability mechanisms for the resourcing of governance institutions and capacity at national and regional levels. - Third, it can provide a legitimate venue for the voices currently absent from both political and technical processes, including the Global South, civil society, and the AI training workforce to shape governance architecture rather than respond to it. The test of success is whether the Dialogue produces outputs that are adopted and implemented. Principles without uptake mechanisms do not advance governance. The Dialogue should set a norm: every session produces at least one concrete, measurable commitment, not only a summary of perspectives.

Several existing initiatives represent genuine foundations rather than parallel tracks. The OECD AI Policy Observatory and its AI Principles, adopted by over 40 countries, provide the most developed international reference point for governance standards. The Dialogue should build directly on this work rather than restart definitional debates, while extending its reach to countries not currently in the OECD network. The Global Partnership on AI (GPAI) has developed substantive technical work on responsible AI, data governance, and AI in the labor market that has not yet been adequately connected to UN-level political processes. The Dialogue should serve as the bridge. The AI for Good Global Summit, co-created with ITU, has since 2017 demonstrated that inclusive, multi-stakeholder AI dialogue is achievable at global scale and that it produces tangible outputs - applications, partnerships, and policy recommendations -when structured well. Its model of connecting technical demonstration with governance discussion is relevant. The Independent International Scientific Panel established under the Global Digital Compact provides the evidence base the Dialogue needs. Its first report should be treated not as background material but as the analytical foundation for every substantive session. The added value the Dialogue specifically brings is universality and political weight. It is the only recent forum where every UN Member State has standing alongside all relevant stakeholders. That legitimacy, if used well, can convert the outputs of existing initiatives into genuinely global commitments. The risk is that it becomes a venue for restating what other processes have already said. The opportunity is that it becomes the place where those outputs are adopted.

Different stakeholder groups bring different and non-substitutable contributions, and the format should reflect this explicitly rather than treating multi-stakeholder participation as a single category. Governments bring political authority and the ability to make binding commitments. Their role is not primarily to share perspectives but rather to make decisions. Governmental segments should be structured around specific commitment points, not general statements. The technical community: researchers, engineers, standards bodies, corporate research laboratories brings the knowledge required to assess whether governance proposals are implementable. They should be embedded in substantive sessions, not confined to side events. Civil society and affected communities bring the perspectives of those who bear the consequences of AI decisions. Their contributions are most valuable when they are structurally integrated into the design of sessions, not invited to respond to agendas set by others. The AI data community, namely data provider, content moderators, and feedback contributors are currently not represented as a stakeholder category in any major AI governance forum. The Dialogue should establish formal standing for this group. Industry brings implementation experience and resources. Their contributions are most credible when they involve specific commitments and technical detail, not only statements of principle. One additional comment on formats: The Dialogue should use problem-focused breakout formats, with outcome documents that name specific commitments by specific actors, not only convergences of view.

Three communities are systematically underrepresented and their absence is a structural problem, not a logistical one. Communities in the Global South that are subject to AI deployment without having shaped the systems being deployed represent a second critical gap. Inclusion here means governance authorship, not consultation, the capacity to set agenda items, propose mechanisms, and veto approaches that do not serve their contexts. This requires investment in participation infrastructure: governance literacy, institutional capacity, and representation mechanisms that function at the local level. Young people and future generations, whose relationship with AI will be longer and more consequential than that of current decision-makers, are structurally underrepresented in bodies organized around current institutional authority. Youth engagement should be built into governance architecture, not treated as an outreach activity. Practical inclusion mechanisms include: supported participation funding for Global South delegates; formal stakeholder categories for labor and worker organizations; co-design of session agendas with underrepresented communities; and outcome documents that explicitly track whose perspectives shaped which commitments.

he formats that work best are those that require specificity and create accountability. On the opposite of the panel-plus-open-floor model that dominates most multilateral dialogue. Problem-centered breakouts work better than thematic panels. Instead of "a discussion on trustworthy AI," a session structured around a specific governance failure or a case of algorithmic harm, a cross-border enforcement gap, an oversight breakdown in an agentic system will generate more concrete and actionable exchange. The Independent International Scientific Panel's findings should be used as the basis for these cases, grounding discussion in evidence rather than position-stating. Commitment tracking in real time creates accountability that summary documents do not. Sessions should include visible tracking of specific commitments made by specific actors: governments, companies, institutions, and with follow-up mechanisms built into the Dialogue's structure from the first session, not retrofitted later. Cross-constituency working pairs: pairing a government delegate with a civil society representative or a technical expert with a practitioner from the Global South on specific problem statements produce outputs that neither constituency would generate alone, and build the working relationships that make implementation more likely.

Several approaches offer concrete lessons for the Global Dialogue. The EU AI Act's risk-tiered approach, calibrating governance requirements to the severity and reversibility of potential harms, provides a practical framework that has moved from principle to regulation. Its limitation is jurisdictional scope; its lesson is that specificity and proportionality make governance implementable in ways that general principles do not. Enterprise-scale Responsible AI governance architecture as developed in large organizations navigating multiple regulatory environments simultaneously demonstrates what operational governance looks like: continuous monitoring, dynamic policy enforcement, chain-of-accountability in multi-agent pipelines, and structured escalation mechanisms. This operational knowledge has not yet been adequately translated into international standards. The Dialogue should create mechanisms for this translation. AI Commons, co-founded in 2017, offers a model for treating AI governance as a global commons problem: where the knowledge underlying AI systems is understood as humanity's shared heritage, and governance frameworks are designed to ensure universal benefit and participation. Its core principle is that every person has the right to benefit from AI and to participate in how it is built and governed, and can provide a foundation for the legitimacy and inclusivity that the Dialogue needs.