MosaicDM

The global community has established a strong foundation of principles around AI. The next step is demonstrating that governance can work in practice. This requires that the systems making consequential decisions on behalf of governments, institutions, and citizens can be held to account before something goes wrong. MosaicDM has spent years building governance infrastructure for AI. This is actual architecture that enforces admissibility before execution: the requirement that an AI system demonstrate deterministic, verifiable compliance with defined parameters before it acts, before it is reviewed, before the damage is done. That distinction is a critical consideration for the Dialogue in Geneva. A successful first Dialogue moves the global conversation from aspiration to enforcement. Voluntary principles have established an important baseline but are not sufficient on their own. Interoperable, binding governance mechanisms are likely to play a central role going forward, and that path requires a common technical baseline. Trustworthy AI loses meaning if every jurisdiction defines it differently and no architecture enforces it consistently. A successful first Dialogue also reckons honestly with who gets left behind. The Global South cannot close the AI divide by inheriting governance models built for economies that can already absorb AI's failures. Capacity-building must mean something more than access to tools never designed with those contexts in mind. It must include governance infrastructure that is operable without the legal and institutional overhead that current frameworks assume, and that respects the data sovereignty of the economies generating that data. MosaicDM is not merely an observer of this problem. We are practitioners who have built the infrastructure this conversation is circling around. One outcome that would signal meaningful success in Geneva is the establishment of the conditions for real governance to take hold globally, with the technical rigor, institutional legitimacy, and human accountability that the moment demands. The Dialogue reflects the high stakes associated with AI governance.

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

MosaicDM builds deterministic AI governance infrastructure. Our selection reflects where that work intersects most directly with the mandate of this Dialogue and where we believe the absence of enforceable mechanisms carries the greatest consequences. Safe, secure and trustworthy AI is the foundation of this work. Our Dimensional Intelligence platform operates on a principle we call admissibility before execution: AI systems must demonstrate verifiable, deterministic compliance with defined governance parameters before they act. Safety and trustworthiness, in our view, are architectural properties, not aspirational ones. Interoperability of governance approaches is one of the areas where global stakes are particularly high. Governance frameworks are proliferating across jurisdictions faster than they can be reconciled. MosaicDM's architecture was designed to function across institutional, national, and regulatory boundaries precisely because fragmentation is the current condition. A Dialogue that produces coherence here creates real infrastructure for international cooperation. Transparency, accountability, and human oversight are the principles our technical architecture is built to enforce. Deterministic systems produce traceable, auditable outcomes. Every decision MosaicDM's platform governs can be interrogated, attributed, and corrected. This is what meaningful human oversight looks like in practice, and it is what the Dialogue should be asking of all AI systems operating at institutional scale. The social, economic, ethical, cultural, linguistic and technical implications of AI ground the other three in consequence. Those implications extend to sovereign nations that are data-rich but lack the institutional infrastructure to govern how that data is used, retained, or monetized. Data sovereignty, the right of communities and nations to control the data they generate, must be a design constraint in any governance framework that claims to be equitable. Taken together, these four areas point to a common conclusion: that governance without enforcement is not governance, and that the Dialogue has the opportunity to establish that standard globally.

Yes. The listed themes address what AI governance should achieve. They place less emphasis on what is required to implement these outcomes. One significant gap is enforcement architecture. The themes speak to transparency, accountability, and oversight as outcomes. They do not address the technical and institutional mechanisms required to produce those outcomes at scale and across jurisdictions. Governance without a deterministic enforcement layer remains advisory regardless of how it is framed politically. The Dialogue would benefit from a dedicated discussion on what enforcement actually requires: technically, institutionally, and in terms of international legal standing. This is a central consideration and a key condition influencing whether governance frameworks hold in practice. A second gap is the treatment of AI governance as financial infrastructure. Data is becoming a collateralizable asset class. When AI-generated data assets underpin financial positions held by sovereign wealth funds, development finance institutions, and central banks, the integrity of the systems producing and managing them becomes a systemic risk question. This area is not fully addressed within the current themes. Economies generating data at scale deserve governance frameworks that recognize data sovereignty and protect the value that data creates, rather than frameworks that allow it to be extracted without recourse. A third gap is energy and infrastructure sovereignty. The assumption embedded in most AI governance discussions is that governance infrastructure requires AI-scale compute. It does not. Deterministic governance architectures that map information geometrically and holomorphically rather than through probabilistic pattern matching operate at a fundamentally different energy profile. This matters enormously for the Global South, where proposals to build large-scale data center infrastructure carry significant energy and resource costs that may be unnecessary given the availability of more efficient governance architectures. The Dialogue could examine whether the infrastructure assumptions embedded in current governance models are necessary or simply inherited. These gaps are connected. Enforcement architecture is what makes governance real. Data sovereignty standards are what make governance equitable. Energy efficiency is what makes governance sustainable for the economies that need it most.

MosaicDM builds deterministic AI governance infrastructure designed for institutional environments across multiple jurisdictions. These gaps reflect conditions that governance architectures are increasingly being designed to address. The absence of deterministic enforcement standards creates asymmetric exposure. Organizations deploying AI in high-stakes environments, financial systems, national security, critical infrastructure, lack a widely recognized international baseline against which to assess whether their governance is adequate. They are making consequential architectural decisions in a standards vacuum. The result is that governance quality varies not by risk level but by the sophistication and resources of whoever built the system. That compounds across borders when those systems interact. The interoperability gap is increasingly significant for organizations operating across jurisdictions. Regulatory frameworks in the EU, the Gulf, Southeast Asia, and North America are diverging faster than any single organization can track. The question is rapidly shifting from which framework to comply with to how to build systems that remain defensible as frameworks continue to shift. The cost of that uncertainty falls disproportionately on institutions in emerging markets that lack the legal and technical resources to maintain compliance across multiple evolving regimes simultaneously. The data sovereignty gap is equally consequential. Nations and institutions that generate significant data have no reliable governance infrastructure to ensure that data is not retained, mined, or used for purposes beyond the immediate transaction. MosaicDM's architecture addresses this as a foundational design principle: we do not retain, mine, or train on the data we govern. For data-rich economies and sovereign actors, that distinction is often a precondition for participation, not a compliance checkbox. The opportunity is significant. Organizations and jurisdictions that establish deterministic, auditable governance infrastructure now will hold a durable advantage as enforcement regimes mature. MosaicDM exists to close the distance between governance as policy and governance as practice. Geneva is where that work becomes a global conversation.

The AI Dialogue arrives at a moment when the gap between governance ambition and governance reality is widening faster than any single jurisdiction can close it. One of its most important roles is to create the conditions under which frameworks produced across different contexts can actually work together. International cooperation on AI governance has stalled repeatedly on the same problem. Every jurisdiction that develops a governance approach does so from its own legal tradition, institutional capacity, and risk tolerance. The result is a landscape of sincere but incompatible efforts, each internally coherent and collectively unenforceable. The Dialogue is one of the first UN mechanisms with the mandate and the membership to address that directly. To do so, it needs to operate at the level of architecture, not just policy. Agreeing that AI should be transparent, accountable, and human-centered is necessary but not sufficient. The Dialogue should be asking what technical and institutional structures make those properties enforceable across borders, and which of the governance approaches already in development come closest to meeting that standard. The Dialogue also has a role that no bilateral or regional process can play: ensuring that the Global South participates in governance design, not just governance adoption. Many of the world's most data-rich economies are also among its most resource-constrained. Current frameworks do not protect data sovereignty or recognize the value that data generates for the communities and nations producing it. A Dialogue that examines governance architectures requiring neither AI-scale compute nor large institutional infrastructure to implement would change the terms of participation for a significant portion of the world. MosaicDM believes the Dialogue's most durable contribution will be establishing that governance interoperability is a technical requirement. That reframe changes what cooperation looks like and what it can achieve.

Several initiatives have laid genuinely useful groundwork. The OECD AI Principles established a working vocabulary for trustworthy AI that 47 countries have now endorsed. The EU AI Act represents the most operationally detailed attempt to translate principles into enforceable requirements. The Bletchley Declaration brought frontier AI safety into intergovernmental conversation for the first time. The G7 Hiroshima AI Process produced voluntary codes of conduct that G7 governments endorsed and encouraged industry to adopt. The African Union's Continental AI Strategy and UNESCO's Recommendation on the Ethics of AI bring non-Western governance perspectives that the broader conversation has been slow to integrate. Each of these efforts has value. None of them alone solves the interoperability challenge, and taken together they represent a fragmentation risk as much as a foundation. The added value the Dialogue can bring is precisely what none of these mechanisms can provide on their own: universality of membership combined with a mandate to connect rather than compete. The OECD reaches its members. The EU regulates its market. The G7 reflects its economies. The Dialogue sits inside the UN system, which means it has the standing to ask how all of these efforts relate to one another and what a jurisdiction outside all of these groupings is supposed to do. MosaicDM would specifically encourage the Dialogue to engage with emerging technical standards work at ISO and ITU, and to examine the infrastructure assumptions embedded in existing frameworks. Governance architecture that operates without AI-scale energy and compute requirements is directly relevant to the Global South, where data sovereignty and sustainable infrastructure are governance prerequisites, not afterthoughts. The Dialogue's unique contribution is coherence across this landscape. That is what the moment needs most.

The Dialogue's credibility will likely depend on whether it produces genuine exchange or managed performance. Format will play a key role in that distinction. Governments have the only authority that can make governance binding. Their role in Geneva should not be to narrate national positions already documented elsewhere, but to negotiate the actual points of disagreement that are preventing interoperability from advancing. The plenary segments should be structured around those points, not around consensus that already exists on paper. The private sector often carries operational knowledge not always available to government delegations. Organizations building and deploying AI at institutional scale know where governance frameworks fail in practice, where enforcement is technically impossible as written, and where compliance theater substitutes for real risk management. That knowledge should be incorporated into the thematic discussions as working input, not merely in side events. Civil society and academia provide the accountability function. They should have standing to challenge, not just to observe. MosaicDM's strongest recommendation on structure is this: the Dialogue would benefit from a practitioner track. A parallel working session where technical and operational experts engage with specific governance problems and report findings back to the main session. The thematic cluster format as proposed risks producing conversations that run alongside each other without ever intersecting. Structured translation from practitioner findings to intergovernmental deliberation is what would make this Dialogue genuinely useful. Representation from the Global South must be substantive. That means structuring sessions around governance problems most acute in data-rich, resource-constrained economies, where data sovereignty and infrastructure sustainability are not peripheral concerns but the central ones. Those voices belong in the practitioner track, not only in the plenary. Without a formal continuity mechanism, each session starts over, and the Dialogue risks becoming an annual conference rather than a cumulative one.

One of the most consequential gaps in global AI governance discussions is not geographic, though geography is part of it. It is the absence of the people who will bear the greatest consequences of AI governance failures and who have the least influence over the frameworks being built. Governments in the Global South are formally present in multilateral processes but substantively peripheral to them. The frameworks under discussion were designed in Washington, Brussels, and Beijing, for economic and institutional contexts that do not reflect the majority of the world's population. Participation without authorship risks falling short of meaningful inclusion. Communities whose data trains the systems that govern them often lack a meaningful seat at the table where those systems are evaluated. Content moderators, data labelers, and the populations whose behavior, language, and cultural production feed global AI models are generating the raw material of this technology without any corresponding voice in its governance. Many of these communities are concentrated in economies that are data-rich but derive little of the value that data creates. Governance frameworks that do not protect data sovereignty, that do not guarantee communities control over whether their data is retained, mined, or used for training, risk being extractive in practice. The Dialogue should address this directly. Small and medium enterprises operating in emerging markets face compliance obligations designed for organizations with legal departments and regulatory affairs teams. Frameworks that cannot be implemented by the organizations they are meant to govern risk functioning more as liability structures than governance tools. They risk functioning as liability documents for the under-resourced. Inclusion requires more than representation in plenary sessions. It requires that the governance problems of underrepresented communities be treated as design constraints. Governance infrastructure built only for the center will fail at the edges. MosaicDM builds for both, and we know the edges are where most of the world lives.

Formats that tend to produce meaningful exchange are the ones that create conditions for disagreement to surface productively. Most multilateral processes are designed to suppress disagreement in favor of consensus language. The AI Dialogue should resist that instinct. One useful innovation would be structured problem sessions built around specific, unresolved governance questions rather than thematic overviews. The question on the table should be: here are three proposed interoperability standards and here is where they conflict, which tradeoffs are acceptable and to whom. That level of specificity can encourage more substantive engagement and generate findings that can actually inform the next session. Practitioner working groups operating in parallel to the plenary, with a formal reporting mechanism back to the main session, could help close the gap between the policy conversation and the operational reality it is meant to govern. The people who know where frameworks break down in practice are rarely the ones at the plenary table. Structured access changes that. Red team sessions, where invited experts are tasked with identifying how proposed governance approaches could fail or be circumvented, would add a rigor that consensus-building processes typically lack. Governance frameworks that have not been stress-tested are not ready to be internationalized. This includes testing whether frameworks can be implemented in resource-constrained environments and whether they protect data sovereignty in practice, not just in principle. For inclusion, asynchronous participation mechanisms that allow organizations and communities without the resources to send delegations to Geneva to contribute substantively to the thematic discussions would broaden the input base without requiring physical presence. This is particularly important for voices from the Global South, where the governance problems are most acute and the representation most thin. One measure of an effective format is whether it changes what gets said, not just who is in the room. Geneva has the opportunity to set that standard for every session that follows.

Some of the most instructive examples of effective AI governance tend to be the ones where enforcement architecture exists alongside principles. The EU AI Act is among the clearest current examples of governance moving from aspiration to obligation. Its risk-tiered approach assigns requirements based on intended use and the potential for harm to health, safety, or fundamental rights. Its limitations are equally instructive: compliance timelines that have outpaced the development of harmonized technical standards, and enforcement mechanisms that assume a level of institutional capacity many deployers do not have. The NIST AI Risk Management Framework represents a different approach: voluntary, flexible, and designed to be implemented across a wide range of organizational contexts. Its value is adaptability across sectors. Its recognized limitation is that voluntary frameworks tend to produce voluntary outcomes. Singapore's Model AI Governance Framework is widely regarded as one of the more operationally useful documents produced by any government on this subject. It translates governance principles into concrete organizational practices with enough specificity to be useful to organizations that are not starting from a policy research background. MosaicDM's Dimensional Intelligence platform represents a practitioner's response to the problem these policy frameworks are trying to solve. Built on the principle of admissibility before execution, it provides deterministic, auditable governance infrastructure that enforces compliance at the architectural level before an AI system acts. It operates without the energy-intensive compute infrastructure that probabilistic AI systems demand. MosaicDM's architecture maps information geometrically and holomorphically rather than through pattern matching, which means it can be deployed in environments where large-scale data center infrastructure is unavailable or unsustainable. This is directly relevant to the Global South, where the push to build AI-scale data infrastructure carries real energy and resource costs that more efficient governance architectures can make unnecessary. MosaicDM also operates on a foundational design commitment: we do not retain, mine, or train on the data we govern. Data sovereignty is a structural property of how the platform is built, not a policy layer added afterward. For sovereign actors and data-rich economies, that is a precondition for trust. A consistent theme across these examples is that governance that cannot be operationalized is unlikely to be consistently practiced. The Dialogue should elevate the examples that close that distance and examine honestly the ones that have not.