Pepperdine University

Governance is supposed to protect people from harm. The people closest to AI's consequences — workers who see systems fail in practice, communities who absorb its outputs, practitioners who know what the data cannot capture — hold knowledge governance needs to do that. In governance processes worldwide, that knowledge does not reliably arrive. Not because stakeholders are absent. Because governance was never designed to receive it. Organizations and governments have built frameworks, adopted principles, and formally seated stakeholders. What they have not built is the structural conditions under which stakeholder knowledge is actively solicited from those who hold it, offered under genuine protection, received with visible confirmation, and given real weight in what gets decided. Inclusion has been treated as the solution. It is not. Without the architecture to receive what stakeholders know, inclusion produces the appearance of accountability without the substance of it — and that gap is where harm accumulates. What makes this failure so durable is that governance cannot see it from the inside. Decision-makers operating within processes never designed to receive certain knowledge do not experience its absence as a gap. They experience sufficiency — the confidence of not knowing what they are missing. That self-concealing quality is precisely what makes it dangerous. It operates at every level of governance worldwide. And it will not be resolved by another set of principles. The first Global Dialogue will succeed if it delivers what no prior effort has: an honest reckoning with the difference between governance that exists and governance that actually works. Name the design failure. Establish shared global standards for governance architecture — not just governance content. Build in mechanisms for ongoing reassessment, because the communities AI affects will not hold still while governance catches up. The world is not short on commitments. It is short on governance designed to hear.

• Safe, secure and trustworthy AI
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Interoperability of governance approaches
• Transparency, accountability, and human oversight

These four priorities are not separate problems. They are four expressions of a shared structural failure - and until that failure is named, progress in any of them will remain incomplete. AI cannot be safe, secure, or trustworthy if the people closest to its consequences cannot get what they know into the processes that govern it. The knowledge that could prevent harm exists. It sits with workers who observe how systems behave in deployment, with communities who experience algorithmic decisions daily, with practitioners who understand context no dataset captures. In governance processes worldwide, that knowledge does not consistently arrive - not because stakeholders are excluded, but because governance was not built to draw it in. Safety without a design for reception remains aspiration. AI's social, economic, ethical, cultural, linguistic, and technical implications cannot be governed from a single vantage point. No governance body - however well-resourced or well-intentioned - holds the knowledge required to anticipate how AI systems interact with communities it has never consulted, in languages it has never worked within, across economies shaped by histories it does not hold. That knowledge is distributed. Governance must be built to reach it. Transparency, accountability, and human oversight require structure behind them to mean anything. A traceable line from stakeholder knowledge to the decisions governance produced - and a return signal to contributors - is what converts oversight from observation to consequence. Without that architecture, transparency is disclosure without accountability and oversight is presence without power. Interoperability is where the global opportunity lives - and where the global risk compounds. If frameworks across jurisdictions align at the level of principles while sharing the same structural inability to receive stakeholder knowledge, interoperability scales the failure. The design question beneath all four priorities is the same: is governance built to receive what it needs, or only what it already expects?

The most dangerous gap in AI governance today is one governance cannot see from the inside: the systematic absence of knowledge it was never built to receive. Every theme on this list - safety, capacity-building, human rights, transparency, interoperability - depends on a structural precondition that none of them names: whether governance is designed to hear. The global conversation has invested deeply in what principles should guide AI, who should participate in governance, and what regulatory frameworks should exist. These are necessary. They also rest on an assumption that has gone unexamined at every level of governance worldwide: that once stakeholders are included, their knowledge will travel - that it will reach deliberation, be recognized as relevant, and carry weight in what gets decided. Our work suggests it does not. Governance processes can be formally inclusive while remaining structurally unable to receive what participants actually know - seating affected communities, soliciting input from frontline workers, convening cross-sector dialogue, and still not hearing. The people governing will not experience this as failure. They will experience sufficiency - the confidence of operating within the limits of what they already hold. That is where harm accumulates, quietly, across every governance context represented in these themes. This is not a values failure. It is not a capacity failure. It is a design failure - and it scales. If international governance frameworks are built without attention to how knowledge moves from affected populations to actual decisions, the same structural gap will reproduce itself globally, theme by theme. The solution is architectural. Governance must be deliberately designed so that stakeholder knowledge is actively solicited, offered under genuine protection, visibly confirmed as received, and given traceable weight in what gets decided. That is the cross-cutting issue beneath every theme on this list - and the one this Dialogue has the authority to name first.

In the United States, AI governance has advanced rapidly in volume — executive orders, federal frameworks, state-level legislation, sector-specific guidance — and more slowly in structural capacity. The challenge is not an absence of governance activity. It is that governance activity has outpaced governance design. Frameworks exist. What remains underdeveloped is the architecture within those frameworks that determines whether the knowledge of affected communities, frontline workers, and non-dominant stakeholders actually reaches the decisions being made. This gap produces two compounding effects. First, governance processes that are structurally unable to receive knowledge from those closest to AI's consequences will continue producing decisions that reflect the perspectives of those who designed them — and the harms that follow will fall disproportionately on communities already bearing the greatest burden of algorithmic decision-making in employment, lending, healthcare, housing, and public services, and often farthest from governance. That pattern is not unique to the United States, but the scale of AI deployment here makes its consequences especially acute. Second, the absence of design standards — standards specifying not just what governance should say but how it must structurally function to receive stakeholder knowledge — means that the proliferation of governance frameworks creates the appearance of progress without its substance. Organizations can adopt frameworks, establish ethics boards, and publish principles while remaining closed to knowledge they are supposed to channel. The distance between governance as adopted and governance as functioning is growing wider, not narrower, precisely because adoption has been treated as the measure of progress rather than reception. The opportunity is equally clear. The United States has the institutional infrastructure, research capacity, and practitioner base to lead on governance design — to move beyond participation rosters and principles toward architectural standards that specify how stakeholder knowledge reaches decisions. What has been missing is the demand. This Dialogue can create it.

Our work suggests the AI Dialogue can do something no prior international effort has managed: shift the global governance conversation from what governance should say to how governance must be built. International cooperation on AI governance has produced significant output — principles, frameworks, guidelines, and resolutions. The next frontier is a shared account of the structural conditions under which those instruments actually function. The EU AI Act and the NIST AI Risk Management Framework each address what governance should require. Neither specifies the conditions under which the knowledge of affected populations reaches the decisions governance produces, even while calling for stakeholder engagement. The gap where governance stalls globally cannot be closed by harmonizing content across frameworks that share the same structural limitation. We believe the AI Dialogue is positioned to play three roles no existing mechanism currently fills. First, the Dialogue can name the design dimension of governance — establishing that how governance receives stakeholder knowledge is as urgent a question as what principles governance adopts. No other forum holds the convening authority or legitimacy to make that shift. Second, it can establish shared diagnostic standards for governance architecture, specifying the structural conditions any governance process must meet for the knowledge of affected stakeholders to be solicited, received, and given weight. This is where meaningful interoperability begins: not alignment of principles, but alignment of structural capacity to hear. Third, it can model what it demands. If the Dialogue itself is structured so that the knowledge of underrepresented stakeholders informs deliberations and carries traceable weight in its outcomes through transparent means, it becomes proof of concept — not a forum for discussion, but a demonstration that governance designed to hear produces governance worth following. The architecture of reception is the missing layer in global AI governance. The Dialogue has the mandate to build it.

The AI Dialogue inherits a dense landscape of governance activity. Its added value lies in addressing what none of it has yet resolved. Several existing initiatives have advanced important dimensions of AI governance. The OECD AI Policy Observatory has built a comparative knowledge base across national approaches. The Global Partnership on AI has convened multistakeholder expertise across working groups. The UNESCO Recommendation on the Ethics of AI established the first global normative framework with near-universal adoption. What these initiatives share — and what the Dialogue is positioned to address — is a common structural gap. Each has advanced governance content: what principles to adopt, what risks to regulate, what rights to protect. None has systematically addressed governance architecture: the structural conditions under which the knowledge of affected stakeholders actually reaches the decisions these frameworks govern. The result is a global ecosystem rich in governance instruments but underdeveloped in the design infrastructure those instruments require to hear. The Dialogue's added value is to build the layer these initiatives are missing. It should connect with and build upon the OECD's comparative work, the GPAI's multistakeholder model, UNESCO's normative foundation, and the growing body of regional governance experience — and introduce a shared diagnostic dimension: whether the governance processes these instruments create are structurally capable of receiving the knowledge they need from the populations they affect. That diagnostic dimension does not exist at the international level today. Existing frameworks assume that stakeholder knowledge will enter governance if participation is invited. Our research and practice suggest otherwise: participation without architecture is presence without influence. The Dialogue has the convening authority, the multilateral legitimacy, and the mandate to establish that distinction at global scale and to build the shared standards that follow from it.

The most consequential design choice the AI Dialogue can make is how its own structure is built — so that what participants know can reach deliberation, carry genuine weight, and be accounted for afterward. Participation without that structure is presence without influence. Three structural recommendations would make the Dialogue's format meaningfully different from prior efforts. First, contribution pathways must be designed, not assumed. Different stakeholders hold different kinds of knowledge: governments hold regulatory experience; civil society holds community impact knowledge; the private sector holds deployment knowledge; academia holds analytical frameworks; technical communities hold system-level understanding; and affected populations hold experiential knowledge no other group can approximate. Each requires a dedicated channel rather than a single open comment period that structurally advantages those with institutional resources and fluency in multilateral discourse. Written input, facilitated sessions, anonymized contribution mechanisms, and asynchronous engagement formats would each reach stakeholders that plenary settings alone will not. Second, visible acknowledgment must be built into the process. Contributors who submit input and receive no signal that it was registered will not trust the Dialogue's process and will not contribute again. The Dialogue should commit to documenting how stakeholder input informed its outcomes and communicating that back to those who contributed, in terms they can access and verify. Third, the Dialogue's outputs should be traceable to the knowledge that produced them. Recommendations, communiqués, and frameworks should be accompanied by an account of which stakeholder inputs shaped them and how. Traceability is what converts a consultative process into an accountable one — and what makes the difference between engagement that informs and engagement that decorates. These are structural design preconditions under which stakeholder participation becomes stakeholder influence. The Dialogue has an opportunity to demonstrate through its own design that governance built to hear produces outcomes worth trusting.

The voices most absent from global AI governance discussions are not unknown. They are structurally unreached — present in the populations AI systems affect most and absent from the processes that govern those systems. Three populations are consistently underrepresented. The first is frontline workers whose daily proximity to AI systems in operation gives them observational knowledge no governance body can replicate from within. These are the people who see how systems actually behave in deployment. Their knowledge is grounded in repeated direct experience of AI's impacts, and it is rarely solicited by governance processes that conflate technical expertise with governance relevance. The second is communities disproportionately affected by AI-driven decisions in employment, lending, healthcare, housing, criminal justice, and public services. Their lived experience of algorithmic outcomes constitutes a form of knowledge that governance processes have systematically failed to receive. Their absence is a design problem: no pathway was built to reach them. These communities have also developed practical ingenuity in navigating — and sometimes redirecting — systems not designed with them in mind. That knowledge is governance-relevant and currently invisible to the processes that need it most. The third is practitioners and civil society leaders in regions where AI systems developed elsewhere are deployed without governance calibrated to local social, economic, cultural, and linguistic realities. Their exclusion is compounded by resource asymmetries, language barriers, and access limitations multilateral formats were not built to overcome. Including these voices requires more than an open invitation. It requires structural design: contribution pathways that reach beyond institutional stakeholders, formats that do not privilege fluency in multilateral discourse, protection mechanisms that make genuine contribution possible for those who face retaliation risks, and acknowledgment processes that confirm input was received and show how it was used. The Dialogue cannot govern communities it has not been built to hear.

The most innovative opportunity before the AI Dialogue is to design a format that actually works — one where the structure of engagement determines that stakeholder knowledge reaches deliberation rather than decorates it. Most multilateral engagement formats are designed for expression, not reception. Plenary sessions, panel discussions, and open comment periods create space for stakeholders to speak. They do not create structural conditions under which what is spoken enters the decision-making process, is visibly acknowledged, or carries traceable weight in what gets decided. The innovation the Dialogue needs is architectural, not procedural. Three format principles would make the Dialogue structurally distinct from its predecessors. Contribution before deliberation. Before any plenary session, structured input should be collected through written submissions, facilitated small-group sessions, and asynchronous digital channels — ensuring that the knowledge of stakeholders who will not speak in large formal settings has already entered the process. Gallery walk formats, where participants record responses visibly before group discussion begins, structurally prevent dominant voices from setting the terms of deliberation before quieter contributors can offer theirs. Real-time acknowledgment. Every contribution channel should include a visible mechanism confirming that input was registered. Contributors who receive silence after offering knowledge will rationally disengage — and the process will lose not only what they offered once but what they might have offered next. Traceable output. The Dialogue's outcomes — recommendations, communiqués, frameworks — should be accompanied by a documented account of which stakeholder inputs shaped them and how. If an outcome cannot be traced to the knowledge that produced it, the engagement process that preceded it was consultative in appearance only. The format the Dialogue adopts will itself be a signal. If it is designed to hear, the world will see what governance built for reception produces.

The most instructive examples of effective AI governance share a single design orientation: they were built not only to include stakeholders but to structurally ensure that what stakeholders know reaches deliberation. At the organizational level, governance processes that composed their decision-making bodies through deliberate, cross-functional selection - rather than defaulting to those already adjacent to technology decisions - consistently accessed a broader range of knowledge. Those who built structured input mechanisms before formal deliberation, rather than relying on voluntary contribution during it, reached knowledge that open discussion formats structurally suppressed. Formats that protected equitable expression - anonymized input channels, facilitated small-group settings, written contributions reviewed before group discussion - consistently surfaced knowledge that open formats suppressed. At the policy level, governance instruments that required documentation of how stakeholder input informed decisions - creating a traceable record from contribution to outcome - produced accountability that disclosure-only approaches did not. Transparency about what was decided is necessary but not sufficient - transparency about how stakeholder knowledge shaped decisions is what converts governance from a compliance exercise into a learning system. What these examples share is the same architectural orientation: stakeholder knowledge was not only invited but designed to arrive. The mechanisms to receive it were built before the knowledge was sought, not assumed to exist because participation was formally permitted. What they also share is the ability to demonstrate that governance designed to hear outperforms governance designed to comply. Where receipt was visible, contribution increased. Where traceability was built in, accountability followed. Where the loop was closed - where contributors learned what became of what they offered - trust accumulated and participation deepened over time. The Dialogue can extend that orientation to the global level. The examples exist. The design principles are transferable. These responses were prepared in collaboration with Dr. Elizabeth M. Adams, Chief Engagement Officer, Minnesota Responsible AI Institute.