HITL Inc.

The first Global Dialogue on AI Governance will be successful if it produces outcomes that change what countries can actually do after the meeting ends. The dialogue should not measure success by whether states agree on broad principles. Most governments have already agreed that AI should be safe, inclusive, transparent, and accountable. The hardest question is whether they have the power, expertise, and institutions to make those principles real. For many developing countries, the AI gap is not just about compute, data or connectivity. It is also about governance capacity. A government may procure an AI system for healthcare, education, security, or public administration without having the technical ability to test it, legal authority to challenge the vendor, or an institutional process to audit its decisions after deployment. In that situation, AI adoption increases dependency rather than closing the divide. The first dialogue should therefore deliver three practical outcomes. First, it should recognise governance capacity as part of AI capacity-building. Second, it should establish a baseline for what countries need to govern AI systems effectively, including audit capabilities, procurement standards, legal authority, incident reporting, and enforceable human oversight. Thirdly, it should recommend practical tools that institutions can use immediately, such as AI procurement checklists, audit templates, risk assessment models, and human supervision requirements for high-risk deployments. The dialogue will succeed if it moves the global conversation from "What should responsible AI mean?" to "Who has the capacity to make AI responsible in practice?" That is the gap the first session should expose and begin to close.

• Transparency, accountability, and human oversight
• Safe, secure and trustworthy AI
• AI capacity-building
• Protection and promotion of human rights

I selected these areas because they directly reflect the work and purpose of HITL, Inc. as a responsible AI governance platform. Safe, secure and trustworthy AI is central to HITL because the platform is designed to reduce the risks of autonomous AI workflows before critical actions are executed. In high-impact environments such as cybersecurity, finance, public services, health, education, national security and critical infrastructure, trust cannot depend only on vendor claims or model documentation. AI systems must be inspectable, auditable, contestable and subject to meaningful human control. AI capacity-building is also a priority for HITL because access to AI tools is not enough. Organisations and governments need the practical ability to evaluate AI systems, define approval points, monitor AI-driven actions, audit decisions and hold vendors or deployers accountable. For developing countries, capacity-building must therefore include governance capacity, not only technical access. Protection and promotion of human rights matter because AI systems increasingly affect people's access to services, opportunities, safety, identity, finance, education, employment and public decision-making. HITL's approach supports rights protection by ensuring that consequential AI actions can be reviewed, challenged or stopped before harm occurs. Transparency, accountability and human oversight are at the core of HITL's work. The platform is built around the principle that human oversight should not only be promised in policy; it should be technically enforced through approval checkpoints, audit trails, escalation routes, named responsibility and the ability to pause or override automated actions. Together, these priorities reflect HITL's contribution to the Dialogue: helping move AI governance from broad principles to practical, enforceable controls in real deployment environments.

The listed thematic areas do not fully capture several cross-cutting and emerging issues. First, implementation capacity and institutional readiness represent a major gap. Most AI governance discussions assume that institutions already possess the legal authority, technical expertise, audit infrastructure, and operational maturity required for effective oversight. In numerous developing countries, this assumption does not hold. Without targeted support for governance capacity beyond general AI access, international frameworks risk being adopted only in name, leading to weak oversight. Secondly, there is a crucial distinction between policy-level commitments and technically enforced human oversight. Declaring that humans should remain "in the loop" is insufficient. AI systems must be built with built-in approval checkpoints, real-time audit trails, escalation mechanisms, and the ability to stop or reject automated actions before they lead to irreversible results. Finally, the dialogue should focus more on high-stakes operational environments (cybersecurity, energy, finance, health, education, law enforcement, and critical infrastructure), where AI failures carry severe real-world consequences. Generic principles often fail to address the practical governance needs of these domains, particularly in ensuring that AI systems are tailored to meet the specific regulatory and operational requirements of high-stakes environments. HITL's work directly supports these issues by enabling enforceable human control in live deployments, helping bridge the gap between high-level policy and practical, auditable implementation, especially for countries and institutions seeking genuine sovereignty and accountability.

These governance gaps affect my sector by exposing the distance between AI adoption and AI accountability. From the perspective of HITL, Inc., the greatest challenge is that organisations are adopting AI agents and automated workflows faster than they are building the controls needed to govern them. In sectors such as cybersecurity, energy, finance, health, education, public services, and national security, AI systems may recommend, trigger, or support consequential actions without a clear process for human review, auditability, escalation, or accountability. The opportunity is to build governance directly into AI implementation. Human approval checkpoints, audit trails, escalation routes and override mechanisms can make AI systems more trustworthy before harm occurs. For developing countries, this approach creates an opportunity to design implementation-first governance models rather than simply copying frameworks from more mature regulatory environments. HITL's work sits within this opportunity. By making human oversight technically enforceable in live AI workflows, HITL helps organisations move from responsible AI as a policy statement to responsible AI as an operational control.

The AI Dialogue can advance international cooperation by helping countries move from shared principles to shared implementation capacity. Many AI governance efforts already agree on broad values, such as safety, transparency, accountability, inclusion and human rights. The most difficult challenge is making these values operational across countries with different levels of technical, legal and institutional readiness. The dialogue can add value by defining the practical conditions required for effective AI governance, including AI-literate regulators, procurement standards, audit mechanisms, cybersecurity capability, incident reporting, vendor accountability and enforceable human oversight. Without these conditions, governance will remain uneven: well-resourced countries will implement safeguards, while less-resourced countries may adopt the language without the tools to enforce it. The Dialogue should also support cooperation through practical resources that countries can adapt locally, such as AI procurement checklists, audit templates, risk assessment models and human oversight standards for high-risk sectors. Its most important role is to prevent a two-tier AI governance system where some countries shape and control AI while others mainly import systems they cannot fully inspect. International cooperation should therefore focus not only on harmonising principles but also on helping all countries build the capacity to govern AI on their own terms.

The Global Dialogue on AI Governance should build upon existing initiatives to ensure coherence and avoid duplication. Relevant foundations include the UNESCO Recommendation on the Ethics of Artificial Intelligence, the OECD AI Principles, the Global Partnership on AI, the EU AI Act, the Council of Europe Framework Convention on Artificial Intelligence, and the ITU AI for Good platform. It should also connect with the Global Digital Compact, the African Union's Digital Transformation Strategy, and relevant UK and European work on AI safety, cybersecurity and critical infrastructure. The Independent International Scientific Panel on AI can provide an important evidence-based anchor for the dialogue. The added value of the UN Global Dialogue lies in its universal and multilateral character. Unlike many existing forums, which are often led by advanced economies or specific regions, the Dialogue can provide developing countries an equal voice in shaping AI governance. Its value should not only be to coordinate existing principles but also to translate them into practical implementation. Specifically, the dialogue can add value by supporting capacity-building governance tailored to the realities of developing countries, including audit capabilities, procurement standards, vendor accountability, incident reporting, and technical skills. It can also promote operational toolkits for high-stakes sectors, such as cybersecurity, finance, health, energy, public services, and national security. From HITL's perspective, the dialogue should place stronger emphasis on technically enforceable human oversight. This means approval checkpoints, audit trails, and escalation mechanisms, as well as the ability to pause or override automated actions before harm occurs. By focusing on these implementation gaps, the Dialogue can help move global AI governance from aspirational commitments to measurable, auditable controls that protect human rights while enabling responsible innovation.

Different stakeholders can contribute distinct forms of knowledge to the AI Dialogue. Governments should share national priorities, regulatory experiences and implementation challenges, especially around institutional readiness, AI sovereignty and public-sector deployment. The private sector and technical community, including governance-focused companies such as HITL, can contribute practical lessons from building and deploying AI systems, including reference architectures for auditability and enforceable human oversight. Civil society and human-rights organisations should bring perspectives from affected communities and highlight risks around exclusion, discrimination and access to remedy. Academia and research institutions can provide independent evidence on risks, capacity gaps and long-term impacts. International and regional organisations can help align the Dialogue with existing frameworks while avoiding duplication. The Dialogue should be structured as a multi-stakeholder and action-oriented process, not only a high-level policy discussion. It should include plenary sessions for broad deliberation, but also focused thematic working groups on implementation issues such as governance capacity-building, technically enforceable oversight and high-risk sector deployment. Regional consultations, especially in Africa, Asia and Latin America, should be used to capture different institutional realities and avoid dominance by a small number of advanced economies. The Dialogue should also include implementation labs or pilot sessions where stakeholders co-design practical tools such as AI procurement checklists, audit templates, human approval checkpoints and accountability frameworks. The process should produce clear deliverables, including operational toolkits, reference architectures and capacity-building modules. It should also include a follow-up mechanism with periodic progress reviews, so that the Dialogue does not become a one-off event. This structure would help bridge policy and practice, strengthen developing-country participation and support safe, sovereign and rights-respecting AI governance.

Developing countries, particularly in Africa, Least Developed Countries, Landlocked Developing Countries and Small Island Developing States, often have limited influence despite being significantly affected by AI deployment. Within these contexts, local institutions, public-sector practitioners, civil society organisations and affected communities are frequently sidelined. This includes rural populations, women, young people, indigenous groups and communities whose languages, needs and risks are poorly represented in mainstream AI systems. Their perspectives on AI sovereignty, institutional readiness, data dependency and context-specific risks are often less visible than those of major technology powers and advanced economies, especially in high-stakes sectors such as public services, health, education, finance, security and critical infrastructure. Operational deployers are also underrepresented. These are the organisations and practitioners integrating AI into real-world systems who understand the gap between high-level principles and technically enforceable controls. These groups should be included through targeted funding, regional consultations, hybrid formats and implementation working groups. Governments, local deployers, civil society, technical experts and governance-focused companies such as HITL should co-design practical tools, including human oversight models, audit templates and accountability frameworks. This would help produce AI governance that is inclusive, practical, rights-respecting and usable in real deployment contexts.

Implementation Labs and Policy-Practice Sprints: Teams made up of governments, tech providers, civil society, and deployers work together to design and test real-world solutions for governance, like human approval steps, audit trails, and ways to handle issues, in areas such as health, finance, or critical infrastructure. These time-bound sprints would produce testable prototypes and toolkits. Regional "Governance Challenges": Country or regional teams present their most pressing implementation gaps (e.g., AI sovereignty or institutional readiness in African contexts) and work with peers and solution providers to develop context-specific approaches. This would amplify under-represented voices from the Global South. Scenario-Based Simulations and Red-Teaming Exercises: Participants engage in realistic high-stakes scenarios (AI failure in public services, autonomous cyber operations, or biased decision systems) to test governance mechanisms in real time. Hybrid Digital Deliberation Platforms: Use AI-supported tools for multilingual translation, real-time polling, proposal ranking, and asynchronous input to enable continuous participation from developing countries and smaller organisations that cannot attend in person. Peer-to-Peer Learning Circles and Mentorship Matches: Pair institutions from mature and emerging AI ecosystems to share practical lessons on capacity-building and enforceable oversight. These formats would encourage active participation, connect policy with real-world technology, and make sure that developing countries and key stakeholders have a say in the results. By prioritising actionable results over declarations, the dialogue can deliver concrete, implementable advances in safe, sovereign, and rights-respecting AI governance.

At the policy level, the EU AI Act provides a useful risk-based model, especially its requirements for high-risk AI systems, human oversight, monitoring, intervention and override. The NIST AI Risk Management Framework also offers a practical approach for identifying, measuring and managing AI risks, with emphasis on transparency, accountability and governance across the AI lifecycle. In developing regions, the African Union Continental AI Strategy and national AI strategies in countries such as Rwanda, Kenya and Egypt highlight the importance of AI sovereignty, capacity-building and context-specific governance. At the practice level, effective AI governance requires more than principles. Human-in-the-loop enforcement should be treated as a technical control, not just a policy commitment. This means approval checkpoints, real-time audit trails, escalation routes and override capabilities before high-impact AI actions are executed. Risk-based governance is also important. AI systems should be classified according to their potential impact, with stronger oversight required in sectors such as cybersecurity, finance, health, energy, public services and critical infrastructure. At the platform level, responsible AI tools such as Credo AI, Holistic AI and Lumenova AI support model risk management, compliance and lifecycle governance. HITL, Inc. contributes by operationalising enforceable human oversight in live autonomous workflows, helping organisations maintain accountability, auditability and control. The AI Dialogue should promote these implementation-focused approaches, support their adaptation to different institutional capacities and encourage open toolkits for technically enforceable governance. This would help close the gap between high-level AI principles and real-world, auditable deployment.