Bath Institute for Digital Security and Behaviour

A successful first Global Dialogue on AI Governance would establish a shared, interoperable framework for addressing AI-related risks and opportunities that crosses regional, cultural, and sectoral boundaries. Success would require moving beyond siloed, sector-specific approaches toward a cohesive, multi-stakeholder governance architecture that is transparent, collaborative, and adaptable to the rapidly evolving AI landscape. Concretely, success would mean agreement on common principles for ethical and responsible AI development - covering safety, accountability, human rights protection, and bias mitigation - alongside mechanisms to ensure these principles are operationalised within real-world development and deployment workflows. The Dialogue should produce actionable commitments rather than aspirational declarations, with clear pathways for implementation and review, connected with standards development processes. A successful outcome would also ensure that the full range of voices is represented, including civil society, academia, end-user communities, and advocates for those without a direct seat at the table, such as marginalised populations and non-human entities referenced in the UN Sustainable Development Goals. The meaningful inclusion of perspectives from the Global South, including small and developing states, would be a critical marker of success. Finally, the Dialogue should establish durable linkages with existing governance mechanisms, including those addressing responsible state behaviour in cyberspace, to build coherence across the broader UN system. The election of H.E. Ambassador Egriselda López as Chair of the Global Mechanism on ICTs and International Security represents an encouraging precedent for this kind of institutional bridge-building. In sum, success lies not in producing a single document, but in launching a living, inclusive, and enforceable governance process capable of keeping pace with technological change while protecting fundamental human rights and societal integrity.

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

These four priorities reflect the most urgent and foundational challenges in AI governance, and their interdependence makes them particularly critical to address collectively. Safe, secure and trustworthy AI is the cornerstone of any effective governance framework. Documented harms, including the use of generative AI for deepfake scams, ransomware development, and the bypassing of content safeguards, demonstrate that safety considerations must be embedded throughout development and operational lifecycles, not applied retrospectively in response to public backlash. Thorough and on-going testing, evaluation of vulnerabilities, and reliable in-built safety features are non-negotiable prerequisites for realising AI's benefits. Interoperability of governance approaches is essential to prevent regulatory fragmentation and the emergence of so-called "avoision" behaviours, whereby organisations exploit ambiguities between avoidance and evasion across jurisdictions. Effective AI governance must function across regions, cultures, providers, and deployers, with sufficient clarity to close gaps that bad actors might otherwise exploit. Protection and promotion of human rights sits at the heart of responsible AI development. The scale and realism afforded by generative AI have substantially amplified existing threats, from targeted disinformation campaigns to discriminatory image generation, with serious implications for dignity, privacy, and equality. Governance must explicitly address those most vulnerable to harm, whether through malicious use or unintended consequence. Transparency, accountability, and human oversight are necessary conditions for public trust. Without clear accountability mechanisms, it is not possible to identify failures, enforce standards, or course correct as capabilities evolve. Human oversight must remain meaningful and not merely nominal, particularly as AI systems are allowed greater degrees of autonomy and likely more consequential in their impacts. Together, these priorities form a mutually reinforcing foundation for governance that is robust, rights-respecting, and fit for purpose across the full spectrum of AI applications.

Several important cross-cutting issues merit explicit attention within the Dialogue's thematic architecture. First, through-life evolution and end-of-life planning for AI systems is insufficiently addressed in current governance discussions. The risks associated with the cessation of an AI system, including dependency, data loss, and the abrupt removal of services that communities have come to rely upon. require dedicated governance consideration, particularly where AI has been integrated into critical public services or infrastructure. Second, the governance of AI at the interface of the natural and non-human world remains largely absent from mainstream discussions. The UN Sustainable Development Goals explicitly recognise the value of life on land and life under water, yet AI governance frameworks rarely consider impacts on biodiversity, ecological systems, or the environmental costs of large-scale AI infrastructure, including energy consumption and hardware disposal. Third, the concentration of AI capability among a small number of large technology companies raises structural concerns about power asymmetries in governance processes. Competing commercial priorities within the technology sector can impede the effective integration of responsible innovation practices, and governance mechanisms must account for the incentive structures that shape industry behaviour. Finally, the psychological and behavioural dimensions of AI interaction - including the ways in which interface design and platform affordances shape user behaviour, invite misuse, or generate dependency, represent an important cross-cutting issue that bridges technical, social, and policy domains. Effective governance requires an understanding of how people engage with AI in practice, not only how systems are designed in principle. These issues cut across all thematic areas and should be incorporated as integral considerations within the Dialogue's agenda rather than treated as peripheral concerns.

N/A

N/A

The AI Dialogue should position itself within, and build upon, the constellation of existing international governance efforts rather than duplicating or fragmenting them. Of particular relevance is the Global Mechanism on developments in the field of ICTs in the context of international security and advancing responsible State behaviour, with which the Dialogue already shares leadership through H.E. Ambassador Egriselda López. The lessons learned from years of UN deliberation on responsible state behaviour in cyberspace, including the challenges of achieving consensus on norms, verification and accountability, are directly applicable to AI governance and should inform the Dialogue's design and ambitions. Collaborative industry-led initiatives, such as Project Glasswing, provide an example of sector-wide approaches to reducing AI-related risks but these should be further expanded to address a broad range of harm types and to include multiple diverse stakeholders and the Dialogue provides an opportunity for this. The Dialogue should seek to build upon such examples, encouraging their expansion in scope and the adoption of similarly collaborative approaches across other harm categories. The OECD AI Principles, the EU AI Act, and the Hiroshima AI Process each offer substantive precedents for translating high-level governance commitments into practical frameworks. The Dialogue should map and synthesise these efforts, identifying areas of convergence that could form the basis of globally applicable standards, while remaining sensitive to the needs of states with different levels of AI capability and regulatory maturity. Finally, the Dialogue should engage with the UN Sustainable Development Goals framework as an anchor for ensuring that AI governance serves the broadest possible conception of human and planetary wellbeing, not merely the interests of the most technologically advanced economies.

N/A

N/A

N/A

N/A