J O M A C A I

In my opinion, Global Dialogue on AI Governance would only be successful if we focus on small, practical and realistic agreements/approaches that help countries work together. Below outcomes I would expect to show genuine success: Common definitions - agree on basic terms like "high-risk AI" so everyone means the same thing. Compatible rules - different countries keep their own laws, but they are shaped so they don't directly clash and are based on shared principles for compliance and oversight. Basic safety standards - shared expectations like reporting AI failures, testing systems before use, and being transparent about powerful models. Cooperation on testing AI - voluntary commitments for sharing methods for checking whether advanced AI systems are safe. Ongoing support for developing countries – proactively helping countries with fewer resources build AI oversight capacity and technical expertise, and access to governance tools. Clear purpose – agreement that it is for coordination, not creating one global law. Next steps – a clear/transparent plan with timelines so work/success continues after the meeting. From my experience, I know that AI governance is too complex and countries have different laws and politics. In my opinion, trying to force one global law would likely to fail. Therefore, I think that a step-by-step, cooperative approach is more realistic, builds trust, and can actually lead to gradual alignment over time.

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

From my experience, the priority is safe AI first, then transparency and accountability, with human rights protected throughout. 1. Safe, secure and trustworthy AI In my opinion, this is the most urgent simply because AI is already widely used. If it is unsafe/poorly controlled, it can cause fast and large-scale harm (e.g. security failures, misinformation, misuse). To me, safety is the base for everything else. 2. Transparency, accountability, and human oversight These ensure AI can be understood, explainable, checked, and controlled. Transparency always helps identify how systems work, accountability ensures ongoing responsibility when things go wrong, and our human oversight allows intervention in high-risk decisions. 3. Protection and promotion of human rights AI must always respect our privacy, promote fairness, and freedom of expression. This ensures technology does not undermine basic rights or deepen discrimination and bias. 4. Social, economic, ethical, cultural, linguistic and technical implications In my opinion, these are important long-term impacts, such as changes to jobs, inequality, and cultural representation, but are less immediate than safety and governance. The priority is safe and controlled AI first, then transparency and accountability, with human rights protected throughout.

Yes, a few important issues are not fully covered: 1. Misuse and security risks - AI can be used for harmful purposes like cyberattacks or scams, so security needs specific focus. Everyone talks about fires safety - why none talks about AI safety. 2. Power concentrated in a few actors - sadly, a small number of people/companies and countries control most advanced AI, which can create unfair influence and access. 3. Environmental impact - AI uses energy and water, create ongoing noise within data centres from generators and humming sound from racks. AI create local grid strains and unplanned power cuts, all of which affects sustainability. 4. Data control and ownership - in my opinion it is important to know who controls the data used to train AI and how/when/where it is used. 5. Fast pace of change - AI is developing faster than any of our rules can keep up, creating ongoing and growing gaps in oversight. In my opinion these issues affect how safe, fair, and effective AI governance is in practice, and they cut across all the main themes listed.

I work globally in the security industry but will focus on the UK context. Key challenges 1. Fragmented regulation - different global approaches (UK, EU, US) create ongoing uncertainty for organisations like ours operating across borders, increasing compliance complexity and slowing innovation/adoption. 2. Limited transparency in AI systems - many organisations incl ours rely on third-party AI tools without full visibility of model design/testing, making audit and assurance processes difficult. 3. Skills gap in AI governance - there is a shortage of professionals who understand both AI systems and security/governance, limiting effective oversight approaches. 4. Infrastructure and resilience pressure – fast growing AI use increases demand on data centres as well as energy systems, raising operational and sustainability concerns. Key opportunities 1. UK leadership in AI assurance - the UK's flexible regulatory approach supports growth in AI auditing, security, and governance services. 2. Expansion of AI security and governance roles ongoing demand is rising for expertise in AI risk, model assurance, and compliance across finance, government, and critical infrastructures. Form my experience in the field, the main challenge is keeping governance and security frameworks aligned with fast AI adoption, but this also creates strong growth opportunities in AI assurance and risk management.

In my opinion, the AI Dialogue can strengthen international cooperation by creating a shared foundation for action on AI governance. It can: 1. Build common understanding – when we agree on key terms like "high-risk AI" and "frontier models" so all countries are aligned. 2. Improve compatibility of rules - help different national approaches unite with shared purpose and work together, reducing regulatory conflict for global organisations. 3. Set basic global expectations – not only encourage but also help implement minimum standards for safety testing, transparency, and human oversight. 4. Support cooperation on advanced AI risks - enable sharing of expertise, knowledge and methods for assessing and managing frontier AI systems. 5. Include all regions - ensure developing countries are part of all discussions and fully supported in building capacity. To me, the Dialogue's role is to encourage, improve coordination and trust between countries, not create one global law, making AI governance more consistent, transparent and practical worldwide.

The AI Dialogue should build on and connect existing global initiatives rather than duplicate them. Existing initiatives to build on: • EU AI Act • GDPR • OECD AI Principles • G7 Hiroshima AI Process • UNESCO AI Ethics Recommendation • NIST AI RMF • UK AI Safety Institute • Asilomar AI Principles Key ISO / standards frameworks • ISO/IEC 42001 – AIMS • ISO/IEC 42005 – AISIA • ISO/IEC 23894 – AIRM • ISO/IEC 25059 – AISQE • ISO/IEC 27001 – AISI Added value of the AI Dialogue • Connect fragmented frameworks, standards and principles into a more coherent global system • Improve interoperability between laws, ethic, cultures, standards, and technical approaches • Translate principles into practice, especially for safety testing and ongoing proactive assurance • Bridge policy, technical, and industry communities globally for mutual understanding and shared purpose • Include wider global participation, especially from developing countries and underrepresented regions/cultures • Create continuity through cross functional working groups and follow-up proactive mechanisms I think that this is an amazing opportunity and the AI Dialogue can act as a coordination hub, linking existing laws, principles, and standards into a more aligned and operational global and transparent AI governance ecosystem.

In my opinion, different stakeholders can contribute in unique ways, and the Dialogue should be structured to make those contributions practical, transparent and ongoing. Perhaps, they should contribute: Governments - to set policy direction, share national/ unique approaches, and work toward alignment on minimum standards (e.g. safety testing, reporting, oversight). Industry Subject Matter Experts (AI developers, deployers, critical infrastructure operators, AI Auditors, AI Ethicists) – to provide technical expertise, share real-world deployment risks, and contribute to safety testing practices, standards and governance framework implementation. Technical and standards bodies (e.g. ISO, NIST-type experts) – to help translate high-level principles into operational standards, metrics, and audit methods as well as to share insights and knowledge. Civil society and academia – to bring independent scrutiny, focus on human rights, ethics, societal impacts, and ensure ongoing accountability and transparency. International organisations (UN system, etc.) – to act as neutral conveners, support inclusivity, and help coordinate global participation, especially from developing countries. I would recommend this format and structure: 1. Thematic small working groups - separate tracks on specific topics: safety, governance interoperability, frontier AI risks, and capacity building. 2. Mixed stakeholder participation - each group should include governments, industry, technical experts, and civil society, to share diverse perspectives and opinions during discussions. 3. Practical outputs, not just statements - each cycle should produce previously agreed tangible deliverables (e.g. shared definitions, testing guidelines, or reporting templates). 4. Technical and policy bridge sessions - dedicated hands-on sessions to translate technical safety work into policy-ready frameworks. 5. Clear follow-up mechanism - regular meetings, clear next steps, progress tracking, and published updates to avoid one-off dialogue with no implementation. In my opinion, the Dialogue should function as a working coordination platform, not just a discussion forum. It should bring all stakeholders together to produce usable outcomes that improve real-world AI governance for everyone.

In my opinion, several important voices are still underrepresented in global AI governance discussions: Underrepresented voices 1. Developing countries - any countries lack resources and technical capacity to participate fully, even though AI systems already affect them through global platforms and services. 2. Frontline workers and affected communities – normal people directly impacted by AI decisions (e.g. in welfare, hiring, policing, or credit scoring) are often not involved in shaping rules. 3. Small and medium-sized enterprises (SMEs) - debates are often dominated by large tech companies, while smaller firms face different challenges in compliance and adoption issues. 4. Technical implementers outside big tech - security teams, auditors, and engineers in government, healthcare, and critical infrastructure are often missing from high-level policy discussions. 5. Civil society from diverse regions - representation is uneven, with limited input and perspectives on ethics, culture, and social impact. How we could include them: • Broaden participation in forums with funded travel and remote access • Create regional dialogue/surveys hubs linked to global discussions • Provide technical and financial support for developing countries • Include "affected communities" panels, not just experts and policymakers • Engage SMEs through dedicated tracks, not general industry sessions • Use structured consultation processes that go beyond one-off meetings In my opinion, a more inclusive AI governance system requires moving beyond governments and big tech. I think that we should actively build space for those who are most affected but least represented.

I would say that, to make the AI Dialogue more than a discussion forum, engagement formats should be practical but interactive and focused on real outcomes. 1. Live "policy + technical" labs - small mixed groups (governments, engineers, auditors, civil society) work on real AI problems together, such as safety testing or incident response. 2. Scenario-based simulations - participants respond to realistic AI crises (e.g. model failure or large-scale misinformation) to identify governance gaps in practice. 3. Global red-teaming exercises - experts collaboratively stress-test AI systems and turn findings into governance and safety recommendations. 4. Regional hubs connected globally - parallel regional sessions linked digitally to ensure broader participation and inclusion of local contexts. 5. Structured public input channels - real citizen panels and consultations to include affected communities, not just experts and policymakers. 6. Short, outcome-focused working sprints - time-limited groups (e.g. 4-6–8 weeks) tasked with producing concrete outputs like shared definitions, standards, or guidance. These hands on formats would make the Dialogue more effective by focusing on collaboration, real-world examples/testing, and producing practical results rather than purely formal discussions.

Below, examples of policies, practices, platforms, and approaches I used and I am familiar with that promote effective AI governance or provide concrete solutions to its challenges: Policies EU AI Act - a risk-based law, approach that classify risks into 4 levels, unacceptable, high, limited and minimal. Why: It reduces harm by focusing regulation where AI can have the most serious impact. GDPR (EU) - data protection law governing personal data use, including in AI systems. Why: It protects privacy and limits unsafe or unfair use of data. NIST AI Risk Management Framework (US) - voluntary framework for managing AI risks across development and deployment. Why: It gives our organisations a practical way to identify and reduce AI risks. Practices AI impact assessments - checking risks before deploying AI systems. Why: Helps identify and prevent harm early. Technical Evidence (Model/System cards, Data Dictionary, Validation Logs) Recording how AI systems are created/trained and what their limits are. Why: Improves transparency and supports auditing. Continuous monitoring - ongoing tracking AI systems after deployment for errors, bias, or drift/changes in behaviour. Why: Ensures ongoing safety, not just one-time checks. Human Oversight: (HIC, HITL, HOTL) - humans either directly approve decisions or supervise and intervene when needed. Why: Keeps human control in high-risk AI decisions. Three Line of Defence- Operational - Managerial - Internal Audit - 1st-day to day users, 2nd - policy makers, 3rd - internal audit team. Why: proactive approach to assign responsibility and accountability at every stage of AI Life cycle. AI red-teaming - testing AI systems by trying to break or misuse them. Why: Finds vulnerabilities before real-world harm occurs. Platforms UK AI Safety Institute testing capabilities - technical infrastructure for evaluating advanced AI systems. Why: Enables independent safety testing of powerful models. OECD AI Policy Observatory - global platform tracking AI governance developments. Why: Helps countries learn from each other and align policies. Hugging Face model hub - platform for sharing and reviewing AI models. Why: Increases transparency and enables public scrutiny. Approaches Risk-based regulation - regulating AI depending on potential harm. Why: Focuses effort on high-risk systems while allowing safe innovation. Multi-stakeholder governance - involving governments, industry, academia, and civil society. Why: Improves balance, legitimacy, and quality of decisions. Technical-policy integration - linking engineering safety work with policy design. Why: Ensures governance is realistic and grounded in how AI works. From my experience I know that effective AI governance works best when laws set clear rules, explainable practices ensure real-world control, platforms increase transparency and approaches continuously connect technical safety with policy-making.