Independent Expert (working for a private sector insurance company)

A successful first Global Dialogue should move beyond high-level principles and produce practical momentum for implementation. Success would include four outcomes. 1) Establish a shared baseline understanding of responsible AI governance across jurisdictions. Countries are taking different regulatory paths. A common language around risk classification, accountability, safety testing, transparency, human oversight, and redress would reduce fragmentation. 2) Create actionable outputs for organisations of different sizes and levels of maturity. Many firms, especially SMEs and institutions in developing markets, need practical guidance rather than complex frameworks. Toolkits, model governance templates, maturity pathways, and capacity-building support would be valuable. 3) Recognise sector-specific realities. AI risks and controls differ across healthcare, finance, insurance, education, public services, and critical infrastructure. The Dialogue should encourage use-case based governance rather than one-size-fits-all rules. 4) Build an ongoing multi-stakeholder mechanism. Governments alone will not solve AI governance challenges. Industry, academia, civil society, standards bodies, and technical communities should remain engaged through working groups and regular progress reviews. A strong first Dialogue would therefore be measured not only by statements made, but by credible next steps, practical cooperation, and a roadmap for implementation.

• Safe, secure and trustworthy AI
• Interoperability of governance approaches
• Transparency, accountability, and human oversight
• AI capacity-building

Safe, secure and trustworthy AI is an immediate priority because trust is foundational to adoption. Without effective controls for security, reliability, testing, resilience, and misuse prevention, organisations and citizens will hesitate to adopt AI at scale. Interoperability of governance approaches is critical because fragmented national or sector rules create complexity, duplication, and barriers to innovation. Greater alignment on core concepts and minimum expectations would help governments and businesses operate responsibly across borders. Transparency, accountability, and human oversight remain essential as AI systems increasingly influence decisions, workflows, and public services. There should always be clarity on who is responsible, when humans intervene, and how affected individuals seek review or remedy. AI capacity-building is equally urgent. Many countries, regulators, SMEs, and public institutions need skills, infrastructure, and governance capability to participate meaningfully in the AI economy. Without capacity-building, the AI divide will widen. Together, these four priorities balance innovation, safety, fairness, and inclusion.

Several important emerging issues deserve greater attention and here are some of the key ones 1) Agentic AI and autonomous systems: As AI tools move from generating outputs to taking actions and decisions, governance needs to address delegated authority, transaction limits, escalation rules, monitoring, and shutdown control. 2) Third-party and supply chain AI risk: Many organisations use AI indirectly through vendors, software platforms, and embedded tools. Governance should cover procurement, contractual accountability, data handling, model provenance, and ongoing continuous assurance. 3) Concentration risk: Dependency on a small number of providers (especially frontier AI). Heavy reliance on limited infrastructure, model, or cloud providers may create resilience, competition, and geopolitical concerns. 4) Environmental sustainability: AI compute demand, water usage, and energy consumption should be considered alongside economic benefits. 5) Assurance and auditability: Independent testing, evidence standards, incident reporting, and measurable controls will be important if governance frameworks are to gain public trust. I believe these issues cut across all sectors and would benefit from coordinated international attention.

In the insurance and wider financial services sector, AI governance gaps are creating both material risks and significant opportunities. A key challenge is uneven maturity across organisations. Large institutions are investing in governance frameworks, controls, and specialist capability, while many mid-sized firms and supply-chain partners remain at earlier stages. This creates inconsistent standards across ecosystems where data, outsourcing, and delegated services are common. Another challenge is the rapid adoption of Gen AI and emerging autonomous tools ahead of established operating models. Many organisations are still developing clear policies for approved use cases, data handling, model risk assessment, human oversight, incident response, and accountability. Where governance lags adoption, risks include data leakage, inaccurate outputs, bias, operational disruption, and regulatory scrutiny. Fragmented regulation is also a growing issue, especially for firms operating across multiple jurisdictions face different expectations on privacy, consumer protection, model transparency, resilience, and accountability. This increases compliance cost and complexity, particularly for global financial markets. At the same time, the opportunities are substantial. In insurance (sector that I am working), AI can improve underwriting quality, accelerate claims handling, strengthen fraud detection, enhance customer service, and reduce manual processing. Better use of AI can also support catastrophe modelling, risk selection, and operational efficiency. For the UK and London market specifically, there is an opportunity to become a trusted global centre for responsible AI adoption in regulated markets. Strong governance standards, proportionate regulation, and practical assurance models would support innovation while protecting customers and market integrity. The most significant need is therefore not simply more AI adoption, but faster development of practical, interoperable governance models that enable safe scaling across sectors.

The AI Dialogue can play an important convening and bridging role by helping countries and stakeholders move from fragmented debate toward practical cooperation. AI development is global, while governance approaches are emerging at different speeds and through different legal systems. The Dialogue can help create common ground without requiring identical regulation. First, it can promote shared baseline principles and terminology on safety, accountability, transparency, human oversight, security, and redress. Common language reduces confusion and supports cross-border trust. Second, it can support interoperability between governance models. Many organisations operate internationally and face overlapping requirements. The Dialogue can encourage mapping between different regulatory and standards approaches so countries retain sovereignty while reducing unnecessary divergence. Third, it can strengthen inclusion. Many developing countries, smaller economies, SMEs, and public institutions risk being rule-takers rather than rule-shapers. The Dialogue can elevate their priorities on access, infrastructure, talent, and capacity-building. Fourth, it can encourage practical cooperation on emerging risks such as frontier AI model safety, cyber misuse, synthetic media, supply-chain concentration, and agentic AI systems. Finally, it can create continuity through working groups, voluntary commitments, case studies, and periodic progress reviews. Its greatest value would be turning broad consensus into usable frameworks, shared learning, and measurable action.

The Dialogue should build upon existing global, regional, and technical initiatives rather than duplicate them. Important foundations include: - Organisation for Economic Co-operation and Development (OCED) AI Principles and policy observatory work - National Institute of Standards and Technology (NIST) AI Risk Management Framework - International Organization for Standardization / International Electrotechnical Commission standards such as ISO/IEC 42001 - European Union AI Act (EU AI Act) The added value of the AI Dialogue is its universal reach and multi-stakeholder legitimacy. Unlike narrower regional or sector efforts, it can connect governments, industry, academia, civil society, and developing economies in one forum. It can add value by: - translating fragmented initiatives into a coherent global map - identifying minimum common governance baselines - highlighting gaps where no forum currently leads - supporting capacity-building for lower-resource countries - developing pragmatic framework for SMEs - sharing practical implementation case studies - encouraging alignment between policy, standards, and real-world deployment Its strongest contribution would be coordination, inclusion, and implementation at global scale.

The AI Dialogue should be structured as an inclusive and action-oriented multi-stakeholder forum where each group contributes distinct expertise. Governments and regulators can share policy approaches, implementation experience, and public interest priorities. Industry can provide operational lessons, technical realities, and case studies on deployment and controls. Academia can contribute independent research and foresight. Civil society can represent rights, labour, consumer, and inclusion perspectives. Standards bodies can help translate principles into measurable practices. Technical and open-source communities can advise on feasibility and innovation pathways. A strong format would combine plenary sessions with practical working tracks. Here are few recommendation to already drafted notes: 1) High-level plenary on shared priorities 2) Thematic roundtables aligned to core UN themes 3) Regional dialogues reflecting different development contexts 4) Sector sessions including healthcare, finance, education, agriculture, and public services 5) Capacity-building workshops 6) Youth and future generations forum The Dialogue would add greatest value if outputs include practical recommendations, shared resources, and follow-up workstreams.

Several important voices remain underrepresented in global AI governance discussions. These include developing countries, small island states, lower-income economies, SMEs, workers affected by workplace transformation, youth, disability communities, indigenous communities, linguistic minorities, and practitioners from highly regulated sectors. Many governance debates are shaped by countries and organisations with greater technical resources. Broader participation would improve legitimacy and produce more practical outcomes. How to include them: - travel support and sponsored participation - remote participation across time zones and where possible use virtual sessions to ensure appropriate participations - multilingual materials and interpretation - regional pre-dialogues feeding into global sessions - open written consultation channels - reserved speaking opportunities - SME and practitioner panels - youth advisory participation defining the need for the future workforce. Inclusive participation is essential if AI governance is to reflect global realities rather than a narrow set of interests.

Traditional panels should be complemented by interactive and solution-focused formats. Few suggestions are: - Scenario labs addressing real governance challenges - Crisis simulations on synthetic media, cyber misuse, or system failures - Policy design sprints for toolkits and templates - Live case clinics sharing implementation lessons - Structured debates on competing approaches - Digital participation platform for polling and collaborative input - Regional hubs linked virtually to the main event - Innovation showcase highlighting responsible AI solutions Outputs should be captured in real time and converted into recommendations or workstreams.

Examples of effective AI governance are emerging across policy, organisational practice, standards, and technical controls. At policy level, the European Union AI Act (EU AI Act) applies a risk-based approach, with stronger obligations for higher-risk uses. The National Institute of Standards and Technology (NIST) AI Risk Management Framework provides a practical lifecycle model. United Nations Educational, Scientific and Cultural Organization has advanced principles on ethics, inclusion, and human rights. At organisational level, strong practices include: - AI inventories or registries - risk-tiering of use cases - cross-functional governance forums - independent review for higher-risk systems - third-party due diligence - post-deployment monitoring - incident response processes - human oversight/human in the loop checkpoints Standards such as ISO 42001 support management system approaches. The most effective models combine innovation enablement with proportionate safeguards and continuous learning loop.