PwC Legal Middle East LLP

A successful first Global Dialogue on AI Governance would not be measured by diplomatic choreography or another carefully worded communiqué reaffirming that AI is important. That much is already settled. Its success would turn on whether it meaningfully advances the legal and institutional grammar of AI governance. First, it should move the debate beyond principle and into design. The real test is whether participants begin to converge around usable regulatory architecture: credible approaches to risk differentiation, obligations across the AI lifecycle, testing and assurance expectations, auditability, traceability, incident reporting, and governance standards for systems with systemic or frontier-level capability. Without that level of specificity, "global dialogue" risks becoming a synonym for managed ambiguity. Second, it should broaden authorship. If the discussion is driven only by major powers and dominant technology companies, it will lack both legitimacy and durability. A genuinely successful Dialogue would create space for emerging markets and institution-building states to shape, not merely inherit, the next generation of governance norms. Third, it should recognise that AI governance is not an ethics sidebar. It sits at the intersection of economic policy, administrative capacity, national security, liability, labour market transition, market concentration, and digital sovereignty. Any serious global forum must reflect that reality. Fourth, success would require continuity. Not endless discussion, but machinery: technical working groups, regulatory exchange channels, common taxonomies, and practical pathways for cross-border coordination. Above all, the Dialogue should help retire one of the laziest binaries in this field: that societies must choose between innovation and regulation. The real task of law is not to suppress technological progress, but to structure it. If this Dialogue helps establish that governance is an enabler of trust, market confidence, and long-term innovation, it will have done something genuinely consequential.

• Transparency, accountability, and human oversight
• Safe, secure and trustworthy AI
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Protection and promotion of human rights

My selected priorities reflect a deeper concern that the development of AI, particularly as systems become more autonomous and agentic, is moving faster than the legal and governance frameworks needed to contain risk in a principled and enforceable way. I am especially concerned that current debates on AI ethics, while important, too often remain at the level of soft principle. As agentic processes develop, capable of initiating actions, making decisions, interacting with other systems, and producing real-world consequences with reduced human intervention, the absence of a sufficiently clear liability framework becomes a serious governance gap. In that context, safe, secure and trustworthy AI is an urgent priority because trust cannot rest on aspiration alone. It requires legally meaningful safeguards, testing, monitoring, and risk allocation. Transparency, accountability, and human oversight are equally critical because, without them, responsibility becomes diffuse at precisely the moment when systems are becoming more capable of acting with practical autonomy. My emphasis on the protection and promotion of human rights reflects the view that these technologies are no longer peripheral. They increasingly shape access to services, economic opportunity, privacy, expression, and dignity. At the same time, the social, economic, ethical, cultural, linguistic and technical implications of AI must remain central, because AI systems do not operate in a vacuum; they reshape institutions, markets, and power relationships across societies. Above all, I am concerned that we may be approaching a point where AI systems are functionally agentic, but our legal frameworks still assume a much simpler model of human control. That mismatch is unsustainable. Urgent action is needed not only to articulate ethical principles, but to develop credible rules on accountability, attribution of fault, oversight obligations, and liability across the AI value chain.

At present, there is no settled global approach to the allocation of liability for outcomes generated by agentic AI systems, and that absence is a matter of serious concern. As AI systems become increasingly capable of acting with reduced human intervention, initiating decisions, interacting with other systems, and producing tangible real-world effects, the traditional assumptions underpinning legal responsibility become progressively less reliable. In particular, there remains significant uncertainty as to how liability should be apportioned across developers, deployers, operators, integrators, and other actors within the AI value chain. This is not merely a theoretical problem. It has direct implications for legal certainty, enforcement, market confidence, and the ability of affected persons to obtain redress. Without clearer frameworks for attribution, causation, duty of care, and accountability, the governance of advanced AI will remain structurally incomplete. A related and insufficiently explored issue is the absence of a coherent framework for AI-related insurance. That gap is likely to become especially consequential in high-impact sectors such as healthcare, where responsible innovation may depend not only on technical safeguards and regulatory approvals, but also on the availability of insurable risk models. In the absence of mature insurance frameworks capable of underwriting AI-related harms, there is a real risk that deployment in critical sectors will either be chilled by uncertainty or proceed without adequate mechanisms for risk transfer and compensation. Taken together, these issues point to a broader regulatory deficiency: AI capability is advancing into increasingly sensitive domains without a correspondingly developed legal and commercial architecture for liability and risk allocation.

In the UAE and wider GCC, the principal challenge is that AI adoption is accelerating far faster than the maturation of the legal and institutional frameworks needed to govern it. The region is rightly ambitious. It is investing heavily in AI as a driver of economic diversification, public sector transformation, digital infrastructure, and geopolitical competitiveness. That creates enormous opportunity, but it also sharpens governance gaps. The most significant challenge is not a lack of policy intent. It is the absence, in many cases, of sufficiently developed frameworks for liability, assurance, transparency, and accountability, particularly as AI systems become more autonomous and more deeply embedded in critical decision-making. This is especially sensitive in sectors such as healthcare, financial services, government administration, and national infrastructure, where the consequences of error, bias, opacity, or system failure are materially higher. A second challenge is institutional readiness. Effective AI governance requires more than strategy documents. It requires regulators, ministries, courts, and supervisory bodies to develop technical fluency, interoperable standards, and operational mechanisms for oversight and redress. In that respect, the governance question is also a state-capacity question. At the same time, the UAE and GCC have a genuine opportunity to lead. Unlike more fragmented jurisdictions, the region is often able to align national strategy, regulation, and implementation with unusual speed. That creates the possibility of building governance frameworks that are not merely reactive, but designed in parallel with deployment. This is also why initiatives such as the UAE General Secretariat of the Cabinet project in the Regulatory Intelligence Office are important. Work on AI-enabled regulatory intelligence and more adaptive legal infrastructure reflects a broader regional opportunity: not only to regulate AI, but to modernise governance itself so that law and regulation can keep pace with technological change.

The AI Dialogue can play a valuable role if it becomes more than a forum for general alignment and instead helps build the conditions for practical international cooperation. First, it can help develop a common vocabulary. One of the current barriers to effective cooperation is that states, regulators, companies, and international institutions are often using the same terms to mean different things, whether in relation to safety, frontier systems, transparency, accountability, human oversight, or risk. A more coherent shared language would itself be a meaningful contribution. Second, it can support convergence around baseline governance principles without requiring full legal harmonisation. That is particularly important in a field where national legal systems, regulatory philosophies, and levels of institutional capacity differ significantly. The aim should not be artificial uniformity, but greater interoperability. Third, the Dialogue can create space for more balanced participation in global rule-shaping. Too often, AI governance is discussed in ways that are dominated by a small number of powerful states or major technology firms. A credible international process should enable emerging markets and institution-building states to help shape the framework, rather than merely adapt to it after the fact. Fourth, it can help identify shared governance gaps that no single jurisdiction can resolve alone. These include questions of liability for agentic AI outcomes, cross-border safety coordination, standards for assurance and incident reporting, and the development of credible insurance and risk-allocation mechanisms for high-impact sectors. Finally, the Dialogue can act as a bridge between principle and implementation by encouraging technical working groups, regulatory exchange, capacity-building partnerships, and ongoing channels of cooperation. Its real value, therefore, lies not in producing another abstract consensus, but in helping build a more operational, inclusive, and internationally usable architecture for AI governance.

The AI Dialogue should build on existing initiatives that have already begun to shape the international AI governance landscape, but it should do so in a way that adds coherence, legitimacy, and institutional continuity. In particular, it should connect with the Bletchley Declaration, which marked an important early moment of international recognition that advanced AI systems present shared risks requiring coordinated global attention, especially in relation to frontier capabilities and safety cooperation. That declaration was significant not only because of its subject matter, but because it demonstrated that geopolitical rivals and diverse states could still converge around at least a minimum common concern regarding AI safety. The Dialogue should also build on the OECD AI Principles, updated in 2024, which remain one of the most developed intergovernmental reference points for trustworthy AI governance and policy interoperability. The added value of the AI Dialogue, however, should not be duplication. Its unique role is to provide a more universal forum under United Nations auspices, anchored in General Assembly Resolution 79/325, capable of bringing together states that may sit outside narrower plurilateral or regional processes. Its real contribution would be to connect these parallel efforts into a more coherent architecture: linking safety discussions such as Bletchley, normative frameworks, standards development, and implementation support. In that sense, the Dialogue can add value by transforming a fragmented field of initiatives into a more inclusive and operational basis for international cooperation.

The AI Dialogue should not be structured as a purely diplomatic event. If it is to have lasting value, it should include a practical institutional component designed to build shared capacity across sectors and jurisdictions. One useful model would be the creation of a multi-stakeholder fellowship programme attached to the Dialogue. Such a programme could bring together a small annual cohort of participants from the public and private sectors, including regulators, government lawyers, judges, standards bodies, technologists, insurers, industry representatives, and civil society experts. Its purpose would be to create a genuinely interdisciplinary community of practice around AI governance, rather than leaving legal, technical, commercial, and policy questions in separate silos. This would be particularly valuable because many of the hardest AI governance questions, especially around liability, safety assurance, accountability, and high-impact deployment, cannot be resolved by lawyers, engineers, or policymakers acting alone. They require structured engagement across disciplines and institutions. In terms of format, the Dialogue could therefore operate on three levels: a high-level plenary for strategic direction, thematic working sessions for focused policy discussion, and a fellowship or lab-style track for deeper cross-sector collaboration. Fellows could be tasked with producing practical outputs, such as model governance tools, comparative policy papers, sector-specific guidance, or recommendations on emerging issues such as agentic AI liability and AI insurance. That would give the Dialogue added value beyond speeches and position statements. It would help create an enduring network of professionals capable of translating international discussion into implementable governance within states, markets, and institutions.

Global discussions on AI governance still tend to overrepresent states, major technology companies, and entrepreneurs building AI-enabled products, while underrepresenting those who will have to apply, supervise, or bear the consequences of these systems in practice. This is especially clear in sector-specific contexts. In healthcare, it is not enough to hear only from founders developing AI diagnostic or clinical tools; the conversation must also include doctors, nurses, hospital administrators, insurers, clinical ethicists, and patients, because they understand where risk actually materialises: patient safety, misdiagnosis, over-reliance, workflow disruption, consent, and liability. The same parallel applies elsewhere. In financial services, governance should not be shaped only by FinTech innovators, but also by compliance officers, prudential regulators, consumer protection bodies, and those responsible for complaints and redress. In education, not only EdTech providers, but teachers, school leaders, and child development specialists should be heard. In the justice system, not only developers of legal AI tools, but judges, advocates, legal aid providers, and court administrators must be part of the discussion. There is also a broader structural imbalance. Emerging markets, smaller states, public sector operators, affected workers, and non-English linguistic communities remain underrepresented, despite often being the most affected by imported systems and externally shaped norms. Inclusion therefore requires more than invitations. It requires deliberate institutional design: funded participation for underrepresented jurisdictions, multilingual processes, sector-based representation, regional consultations, and mechanisms that allow affected professionals and communities to shape outcomes rather than merely react to them. Without that, AI governance risks being overly informed by those who build the technology and insufficiently informed by those who must live with its operational, legal, and human consequences.

To foster meaningful engagement, the AI Dialogue should move beyond a traditional conference format built around formal statements and panel discussions. A more effective approach would combine diplomatic dialogue with structured, problem-solving formats that force interaction across disciplines, sectors, and jurisdictions. One useful format would be issue-specific policy labs focused on concrete governance problems, such as liability for agentic AI, AI assurance in healthcare, cross-border incident reporting, or transparency obligations for high-impact systems. These labs should bring together lawyers, regulators, engineers, insurers, sector specialists, and civil society actors to work through real governance scenarios rather than speak in abstractions. A second valuable format would be cross-sector simulation exercises. For example, participants could be asked to respond to a hypothetical AI system failure in a hospital, financial institution, or public authority. This would expose how different stakeholders perceive risk, responsibility, escalation, and redress, and would make governance gaps far more visible than a conventional panel. The Dialogue could also include a multi-stakeholder fellowship or practitioner track, bringing together a curated cohort from the public and private sectors, including regulators, judges, government lawyers, industry actors, standards bodies, and technical experts. That would help build an enduring interdisciplinary community rather than a one-off exchange. Another strong format would be sector roundtables designed to ensure representation from those directly affected by deployment. In healthcare, for example, discussions should include doctors and hospital operators, not only entrepreneurs building AI-enabled tools. The same principle should apply across finance, education, and justice. Finally, the Dialogue should invite short written provocations or challenge papers in advance, so sessions begin from concrete propositions rather than general remarks. The most effective format, therefore, is one that treats the Dialogue not as a stage for speeches, but as a working forum for institutional design.

A number of existing approaches offer useful building blocks for effective AI governance. First, the OECD AI Principles, updated in 2024, remain one of the strongest intergovernmental reference points because they combine trustworthiness, human rights, accountability, and innovation in a form that is broad enough for international use but concrete enough to guide national policy. Second, the Council of Europe Framework Convention on AI, opened for signature in September 2024, is especially important because it moves beyond soft ethics into a binding legal instrument anchored in human rights, democracy, and the rule of law. That is a significant development for jurisdictions seeking more legally grounded models. Third, from an operational perspective, the NIST AI Risk Management Framework is a strong example of a practical governance tool. Its structure, centred on governing, mapping, measuring, and managing risk, helps organisations translate abstract principles into internal controls, testing, documentation, and oversight processes. Fourth, ISO/IEC 42001 is valuable because it provides a formal management system standard for AI. This is particularly useful for organisations that need governance to become auditable, repeatable, and embedded in enterprise processes rather than handled through ad hoc ethics policies. More broadly, effective governance is likely to come from combining these layers: legal frameworks, risk management practices, technical standards, and institutional capacity-building. In my view, the most promising approach is not a single instrument, but an integrated model in which high-level principles are connected to enforceable rules, organisational controls, and sector-specific assurance mechanisms. That is the direction in which global AI governance now needs to move.