Asahi Intecc USA

Success for the first Global Dialogue on AI Governance would require moving beyond consensus on principles toward the foundation of actionable, enforceable global standards — particularly for AI systems operating in high-stakes domains. Three concrete outcomes would signal genuine progress: First, a commitment to sector-specific governance tracks. General-purpose AI governance frameworks cannot adequately protect patients, workers, or citizens when applied without adaptation to high-risk domains. A successful Dialogue would produce a mandate for dedicated working groups — particularly for AI in healthcare, critical infrastructure, and financial systems — where the consequences of algorithmic failure are irreversible. Second, a harmonization roadmap for regulatory frameworks. Today, AI medical devices, hiring algorithms, and credit systems face entirely different legal obligations depending on the country in which they are deployed. Patients in over 150 countries receive AI-assisted diagnoses with no transparency protections whatsoever. A successful Dialogue would produce a timeline and process for developing a Global Minimum Standard — a floor of transparency, accountability, and human oversight requirements that all member states commit to meeting, adapted to their capacity levels. Third, meaningful inclusion of the technical community in governance design. Policy crafted without practitioners who build, audit, and regulate AI systems will produce frameworks that look coherent on paper but fail in implementation. A successful Dialogue would formalize channels for regulatory professionals, engineers, and civil society technologists to contribute to governance design — not just comment on it. The Dialogue will have succeeded if delegates leave Geneva with a shared definition of what responsible AI deployment looks like in life-critical contexts, a working structure to develop binding minimum standards, and a credible process for including countries currently without any AI regulatory infrastructure in that conversation.

• Safe, secure and trustworthy AI
• Interoperability of governance approaches
• Protection and promotion of human rights
• Transparency, accountability, and human oversight

My selections reflect the priorities of a regulatory professional working daily at the intersection of AI governance and medical device compliance across multiple jurisdictions. Safe, secure and trustworthy AI is foundational. In healthcare, "trustworthy" is not a soft value - it is a clinical prerequisite. An AI system that a physician cannot interrogate or a patient cannot understand is not safe, regardless of its technical performance metrics. Trustworthiness must be operationalized through mandatory disclosure requirements, not left to voluntary best practices. Transparency, accountability, and human oversight is where current international frameworks diverge most sharply - and where harmonization is most urgent. The U.S. FDA, EU, UK MHRA, Health Canada, TGA, and IMDRF each define algorithmic transparency differently. Clinicians and patients bear the consequences of that inconsistency. A global governance framework must establish minimum transparency obligations: documented data lineage, explainable outputs at the point of care, and clear accountability chains when AI-assisted decisions cause harm. Interoperability of governance approaches addresses the structural problem beneath all others. A medical AI device approved under one jurisdiction's framework may carry no equivalent obligations in another. Without interoperability, regulatory arbitrage becomes the default - developers optimize for the most permissive market, and patients in lower-capacity countries receive AI systems with no accountability infrastructure. The Dialogue must prioritize mutual recognition pathways and shared baseline standards. Protection and promotion of human rights grounds the technical conversation in what is ultimately at stake. When AI makes or influences decisions about a person's diagnosis, treatment, or access to care, that person has a right to know. Informed consent for AI-assisted medical decision-making is not merely a regulatory compliance question - it is a human rights issue that governance frameworks have not yet adequately addressed.

Two significant gaps are not adequately captured by the listed thematic areas: 1. Regulatory coverage voids for AI in healthcare in low- and middle-income countries (LMICs) Over 150 countries currently have no functional regulatory framework for AI-enabled medical devices. This is not a capacity-building gap alone - it is a governance architecture failure. AI systems validated on datasets from high-income populations are being deployed in clinical settings in LMICs without any requirement for local validation, bias assessment, or transparency disclosure. Patients in these settings receive AI-assisted diagnoses with no legal recourse, no disclosure rights, and no assurance that the algorithm was trained on data that represents them. The Dialogue should explicitly name this as a governance emergency distinct from general AI capacity-building. It requires not just technical training but immediate adoption of minimum regulatory standards - possibly through WHO and IMDRF - before further deployment scales. 2. Algorithmic transparency as patient consent Current governance discussions frame transparency primarily as an accountability mechanism between developers and regulators. What remains underaddressed is the patient-facing dimension: the right of an individual to know that an AI system was involved in a medical decision affecting them, and to understand its basis. This is not a niche concern. AI is actively embedded in diagnostic imaging, clinical decision support, surgical robotics, and chronic disease management. No existing international framework establishes a patient's right to AI disclosure in healthcare as a baseline. The AI Dialogue should examine whether existing human rights instruments - including the right to health, the right to information, and principles of informed consent - create an existing legal foundation for this right, and what governance mechanisms would be needed to enforce it globally. These two issues cut across healthcare, human rights, and governance architecture in ways none of the listed themes fully captures.

My perspective is grounded in the United States medical device sector, where I work as a Regulatory Affairs Specialist managing AI/ML Software as a Medical Device (SaMD) submissions and compliance strategy. The most significant challenge is regulatory fragmentation creating unequal patient protection. The U.S. FDA has made meaningful progress — the AI/ML-Based SaMD Action Plan, Predetermined Change Control Plans, and the 2024 draft guidance on transparency represent genuine advancement. But these frameworks exist in isolation. A device my organization submits under FDA oversight may face entirely different transparency and accountability requirements if deployed in the EU under the AI Act's high-risk classification, or in markets where no equivalent standard exists. The practical result is that compliance teams spend disproportionate resources navigating jurisdictional inconsistency rather than improving the safety of the technology itself. The opportunity cost is significant. Regulatory uncertainty suppresses responsible innovation. Smaller SaMD developers — particularly startups building AI for underserved clinical conditions — often cannot afford multi-jurisdictional compliance and default to the most permissive available market. This creates a race-to-the-bottom dynamic that harms patients globally and disadvantages developers committed to rigorous standards. The emerging challenge is continuous learning algorithms. FDA's adaptive AI guidance is still evolving. Devices that update their own models post-deployment create accountability gaps that current frameworks — built around static device approval — were not designed to address. No international consensus exists on how to govern AI that changes after it reaches the patient. The opportunity the Dialogue represents is precisely this: the U.S. regulatory framework, while advanced, cannot solve global fragmentation unilaterally. A multilateral process that establishes baseline interoperability — even beginning with mutual recognition of core transparency requirements — would reduce compliance burden for responsible developers while extending meaningful protections to patients currently outside any governance framework.

The AI Dialogue occupies a unique position that no existing multilateral body currently fills: a neutral, inclusive platform with universal membership where governments, technical practitioners, civil society, and the private sector can develop shared governance language without the constraints of trade negotiations or bilateral regulatory agreements. Its most valuable role is not to replicate work already underway in the OECD, G7, or regional bodies — it is to function as the connective layer between those initiatives and the majority of the world that has not yet been part of that conversation. Concretely, the Dialogue can advance international cooperation in three ways: First, establishing shared definitions. Terms like "transparency," "explainability," "high-risk AI," and "human oversight" mean different things in different regulatory frameworks. Before harmonization is possible, common vocabulary is necessary. The Dialogue is positioned to convene the technical and regulatory communities needed to produce definitions that translate across legal systems. Second, creating a mutual recognition pathway for AI governance standards. Full harmonization is neither realistic nor desirable in the near term — regulatory systems reflect different legal traditions and social values. But mutual recognition of core baseline requirements is achievable. The Dialogue should work toward a framework where a jurisdiction that meets defined minimum standards for AI transparency and accountability receives reciprocal recognition, reducing duplicative compliance burdens. Third, giving voice to countries currently outside AI governance architecture. The nations with the least regulatory capacity are absorbing AI systems — including medical AI — at the fastest rates. The Dialogue must ensure their participation is substantive, not symbolic. Governance designed without them will fail to protect their populations and will lack the legitimacy needed for global adoption. The Dialogue's added value is legitimacy through universality — something no regional or club-based initiative can provide

Several existing mechanisms provide important foundations the Dialogue should connect with rather than duplicate: IMDRF (International Medical Device Regulators Forum) has produced the most substantive cross-jurisdictional guidance on AI/ML SaMD to date. Its framework for risk categorization and evidence generation provides a technical baseline the Dialogue should formally reference when addressing AI in healthcare. IMDRF's limitation is membership — it includes major regulators but excludes most of the world. The Dialogue can extend IMDRF's technical work into a broader governance commitment. The OECD AI Principles and the Global Partnership on AI (GPAI) established early consensus on values — human-centered, transparent, accountable AI. These principles have been adopted by over 40 countries. The Dialogue should treat them as a floor, not a destination, and focus on translating principles into measurable, jurisdiction-specific implementation standards. The EU AI Act represents the most comprehensive binding AI governance framework currently in force. Its high-risk classification system, conformity assessment requirements, and transparency obligations for AI in medical devices provide a regulatory model the Dialogue can learn from — including its limitations and implementation challenges, which are already surfacing. WHO's guidance on AI in health and the ITU's AI for Good platform address the development dimension that purely regulatory bodies often miss. The Dialogue should build explicit bridges to both, ensuring health equity and digital inclusion remain central to governance design. What the Dialogue adds that none of these provide is a single, UN-anchored process with universal participation and the political legitimacy to produce commitments that member states will incorporate into national frameworks. OECD reaches wealthy nations. GPAI is advisory. IMDRF is technical. The AI Act is regional. The Dialogue is the only mechanism positioned to produce a globally legitimate governance architecture — if it chooses to be ambitious enough.

Effective stakeholder contribution requires structure that matches the type of knowledge each group holds. The AI Dialogue should resist the default model of plenary statements and side events, which systematically advantage well-resourced delegations and produce declarations without implementation pathways. Governments bring legal authority and political commitment — their role is to translate Dialogue outcomes into national regulatory action. Their contribution is most valuable when focused on identifying specific domestic barriers to implementation, not restating positions already known. The technical community and regulatory professionals hold operational knowledge that governance design cannot afford to ignore. This group — engineers, regulatory affairs specialists, clinical informaticists, standards bodies — should be embedded in working groups, not limited to observer status. Governance frameworks that practitioners find unworkable will not be implemented regardless of political endorsement. Civil society and patient advocates provide accountability and ground-truth on how AI systems actually affect people. Their contribution is most meaningful when they can respond directly to proposed standards — not after frameworks are finalized. Private sector participation should be structured to separate implementation expertise from lobbying interest. Companies that build AI systems have essential technical knowledge; they also have commercial incentives that can distort governance outcomes. The format should distinguish between technical input sessions and policy negotiation tracks. Structural recommendations: The Dialogue should establish standing thematic working groups between its 2026 and 2027 sessions — not just convene twice and issue summaries. Working groups on high-risk AI sectors (healthcare, criminal justice, critical infrastructure) should include mandatory multi-stakeholder composition with defined deliverables. Submissions like this one should feed into publicly accessible synthesis documents that working groups are required to respond to — creating accountability for whether diverse input actually shapes outcomes rather than being received and filed.

The most consequential gap in global AI governance is not ideological — it is geographic and professional. Clinicians and frontline healthcare workers in low- and middle-income countries are among the least represented voices in AI governance discussions, yet they are increasingly the end users of AI diagnostic tools, clinical decision support systems, and AI-assisted surgical devices. Their experience — what these systems get wrong, where they fail, how patients respond — is irreplaceable evidence for governance design. They are absent because participation requires resources, institutional affiliation, and English-language fluency that many lack. Regulatory professionals outside major economies face this same barrier. Countries without mature AI regulatory frameworks are not without knowledgeable professionals — but those professionals lack the platforms, funding, and networks to participate in Geneva-level dialogues. Their absence means governance frameworks get designed for contexts they already inhabit, not contexts where new frameworks are most urgently needed. Patients and communities directly affected by AI-assisted medical decisions — particularly in the Global South — have no formal channel in any current international AI governance process. This is a structural omission with human rights implications. How to include them: The Dialogue should establish a dedicated fellowship or funded participation track for regulatory professionals, clinicians, and civil society representatives from LMICs — modeled on similar mechanisms in climate and health governance. Participation cannot be meaningful if it depends on self-funding. Finally, translation infrastructure matters. Governance that only functions in English is governance designed for a subset of the world it claims to represent.

The formats that have historically dominated UN governance processes — plenary speeches, negotiated communiqués, side panels — are well suited to registering positions. They are poorly suited to resolving technical complexity, building genuine consensus, or producing governance frameworks that practitioners will actually use. Three formats would meaningfully improve the Dialogue's impact: Regulatory simulation exercises. Before each session, participating delegations and technical stakeholders should engage in structured scenario exercises — for example, how would each jurisdiction's current framework handle a specific AI failure event, such as a bias-related misdiagnosis in a deployed medical device? Simulation forces engagement with operational reality rather than abstract principles, surfaces genuine gaps in existing frameworks, and builds shared understanding across delegations that have never had to apply each other's standards. Red-team panels with response obligations. Rather than only convening panels of advocates for governance proposals, the Dialogue should structure sessions where proposed standards are formally challenged by practitioners, ethicists, and representatives of affected communities — and where proponents must respond substantively. This is how standards bodies and regulatory agencies develop durable frameworks. It is largely absent from intergovernmental AI governance. Persistent digital participation infrastructure. The Dialogue should not exist only during its July 2026 and 2027 convenings. A structured online platform — with moderated thematic threads, public comment periods on draft outputs, and real-time translation — would extend meaningful participation to stakeholders who cannot travel to Geneva or New York. The inputs collected through this form suggest appetite exists; the infrastructure to channel that input continuously does not yet. The underlying principle is that format determines who shapes outcomes. If the Dialogue adopts formats designed for diplomatic declaration, it will produce diplomatic declarations. If it adopts formats designed for technical governance, it has a chance to produce frameworks that protect people.

The most effective AI governance approaches share a common feature: they are specific enough to be implemented, flexible enough to accommodate evolving technology, and grounded in the operational realities of the sectors they govern. Several examples are worth highlighting for the Dialogue's consideration. FDA's Predetermined Change Control Plan (PCCP) framework represents a meaningful governance innovation for continuously learning AI systems. Rather than requiring re-approval every time an algorithm updates, the PCCP requires developers to define in advance what changes are permissible, under what conditions, and with what evidence. This approach balances regulatory oversight with the reality that AI improves through deployment - and offers a model other jurisdictions could adopt or recognize reciprocally. The EU AI Act's risk-tiered classification system demonstrates that proportionate governance is achievable at scale. By anchoring regulatory requirements to the severity of potential harm rather than applying uniform obligations, it avoids both under-regulating high-stakes applications and over-burdening low-risk innovation. Its implementation challenges - particularly for SMEs and developers in countries without equivalent legal infrastructure - are instructive for what a global minimum standard should anticipate and address. IMDRF's SaMD framework remains the strongest example of regulators from multiple jurisdictions producing shared technical guidance without requiring full legal harmonization. Its process - technical working groups with defined membership, public comment periods, and iterative drafting - is a governance model the Dialogue should replicate for AI-specific instruments. Regulatory simulation as a governance tool is emerging in practice. Platforms that allow developers and regulators to test submissions against multiple jurisdictional frameworks before filing reduce compliance burden, surface harmonization gaps in real time, and generate evidence about where standards diverge most consequentially. Institutionalizing this kind of pre-competitive regulatory intelligence infrastructure could accelerate convergence across frameworks globally. These approaches share what effective governance requires: clarity, proportionality, and feedback mechanisms that improve the framework over time. This closes the submission on a strong, forward-looking note - and the regulatory simulation reference is a subtle but credible nod to ReguTron's value proposition without being promotional.