Charles Sturt University

Three outcomes may indicate that the first Global Dialogue on AI Governance has been productive. The first concerns normative convergence. Participating states and entities may benefit from developing a shared set of governance principles with sufficient specificity to inform domestic regulatory frameworks. Broad affirmations of intent tend to carry limited policy traction, and more granular agreements on measurable governance expectations may prove more durable across jurisdictions. The second concerns institutional continuity. A standing coordination mechanism in which the Global South holds substantive rather than token representation may help sustain the political momentum generated by the Dialogue. Governance gaps in AI tend to be compounded by venues that underrepresent states most exposed to AI-related harms and least involved in shaping the technology's development. A permanent or semi-permanent multilateral forum could address some of these structural asymmetries over time. The third concerns implementation capacity. Normative agreements tend to have limited effect when a significant proportion of UN member states lack the technical infrastructure, regulatory expertise, and data ecosystems to act on them. A funded capacity-building agenda with accountability structures to track progress may help close the gap between commitments made at the international level and governance realities at the national level. Progress across these three areas may suggest that the Dialogue has moved beyond a symbolic exercise toward a more durable contribution to global AI governance. The degree to which any resulting framework proves equitable and responsive to diverse implementation contexts will likely depend on whether financing and representation commitments are honoured in subsequent processes.​​​​​​​​​​​​​​​​

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

My selections reflect thematic areas where my research and institutional practice have generated direct and transferable evidence. Work on AI detection tools in higher education has illustrated how AI systems deployed in consequential contexts can cause measurable harm when accountability mechanisms are absent. Research published in the Journal of Higher Education Policy and Management found that widely adopted detection tools produce outcomes that are neither reliable nor equitable, raising questions about what safe, secure and trustworthy AI governance requires in practice. This experience informs my view that trustworthiness requires verifiable accountability structures, not affirmations of intent. Transparency and human oversight are closely related concerns in my work. The S.E.C.U.R.E. GenAI Use Framework and the AI Inherent Risk Scale were developed in response to governance environments where AI was being integrated into high-stakes decisions without adequate disclosure to affected parties or meaningful human review. These tools have been adopted across multiple jurisdictions, which suggests the demand for practical transparency infrastructure is broad. Interoperability of governance approaches is an area where my multi-jurisdictional experience across Australia, the United Arab Emirates, the United Kingdom, the United States, and Ireland has been instructive. Governance frameworks developed in one regulatory context tend to transfer poorly to others, and the absence of shared reference points creates conditions where institutions default to locally convenient rather than defensible practices. AI capacity-building underlies all three of the above priorities. My experience as Academic Lead for Artificial Intelligence at Charles Sturt University suggests that governance commitments at the institutional and international level tend to outpace the technical and regulatory capacity of those expected to implement them. Sustained investment in capability may therefore be a precondition for the other thematic areas producing meaningful outcomes.

When AI systems make consequential decisions about people, a governance gap emerges that the listed themes do not yet address with sufficient precision. Research on automated detection tools in educational settings found that these systems produce unreliable and inequitable outcomes, causing documented harm to individuals who had not violated any policy. The same structural problem tends to appear wherever AI operates in enforcement or adjudicative roles. Existing transparency and accountability frameworks largely assume AI in a productive or advisory capacity, and may need extension to cover contexts where the consequences of error are borne by those with least power to contest them. A related gap concerns the distance between governance commitments made at the international level and the institutional conditions needed to act on them. Experience developing the S.E.C.U.R.E. GenAI Use Framework and the AI Inherent Risk Scale suggests that practitioners in organisations with limited regulatory expertise tend to reach for available tools rather than defensible ones. Capacity-building agendas tend to address technical skill acquisition, which covers part of this problem. The more persistent difficulty is the absence of sector-specific guidance, worked examples, and feedback mechanisms that allow normative commitments to become operational practice. Implementation support may therefore warrant treatment as a governance priority in its own right.

In Australian higher education, governance gaps across the selected thematic areas are producing concrete and observable effects. The absence of shared standards for trustworthy AI has meant that institutions have adopted AI tools, including AI detection systems, without adequate evidence of their reliability or equity impacts. Research published in the Journal of Higher Education Policy and Management found that detection tools widely used across Australian universities produce outcomes that are neither accurate nor consistent, and that the harms fall disproportionately on students from non-English-speaking backgrounds. This is a direct consequence of deploying AI in high-stakes contexts before governance infrastructure exists to evaluate or constrain it. The interoperability gap is particularly visible in a sector that operates across multiple jurisdictions and regulatory frameworks. Australian higher education providers deliver programs in partnership with institutions and regulatory bodies across Asia, the Middle East, and Europe. Governance approaches that do not translate across these contexts create conditions where institutions default to the least demanding available standard rather than a defensible one. Transparency and human oversight requirements remain inconsistently applied. Where they exist, they tend to address disclosure at the point of AI development rather than at the point of deployment and use, which is where most institutional decisions are being made. The opportunity in this context is that higher education has characteristics that make it a productive site for governance development. It operates across jurisdictions, serves diverse populations, involves high-stakes decisions about individuals, and has existing quality assurance infrastructure that could be extended to cover AI governance. Frameworks developed in this sector, including the S.E.C.U.R.E. GenAI Use Framework and the AI Inherent Risk Scale, have already demonstrated transferability across national contexts, suggesting that sector-based governance development may generate models with broader applicability.​​​​​​​​​​​​​​​​

The AI Dialogue may be most useful as a mechanism for establishing shared reference points rather than binding agreements. International cooperation on AI governance is currently fragmented across bilateral arrangements, regional frameworks, and sector-specific initiatives that do not communicate well with each other. The Dialogue could contribute by mapping where normative convergence already exists and identifying where divergence reflects genuine differences in values or regulatory context rather than simple coordination failure. A practical contribution would be the development of a shared vocabulary precise enough to support cross-jurisdictional policy comparison. Much existing international discourse uses terms like transparency, accountability, and trustworthiness in ways that are not operationally equivalent across national contexts. This makes it difficult to assess whether different jurisdictions are pursuing compatible or incompatible approaches. Experience developing practitioner-facing governance tools for higher education suggests that shared classification language, even at a relatively general level, can meaningfully improve coordination across institutional and regulatory contexts. The Dialogue may also be positioned to surface governance experiences from sectors and regions underrepresented in existing international forums, including higher education, where AI deployment in assessment and integrity contexts has generated evidence about governance failure with broader applicability.

The Dialogue should engage with UNESCO's Recommendation on the Ethics of AI, the OECD AI Principles, and the Council of Europe's Framework Convention on AI, which represent the most developed existing normative instruments. These frameworks have established baseline language through meaningful multilateral deliberation, and the Dialogue may add more value by building on them than by producing parallel commitments. At the sectoral level, initiatives such as the EDSAFE AI Fellowship and the work of the International Center for Academic Integrity have generated practitioner-facing governance tools with demonstrated transferability across national contexts. These represent a category of implementation-oriented work that intergovernmental processes tend to produce less readily. The Dialogue could add value by creating formal linkages between normative frameworks developed at the international level and practitioner tools developed at the sector level, since this connection between principle and practice is where international governance efforts most often lose traction.​​​​​​​​​​​​​​​​

Meaningful stakeholder contribution requires structural conditions that most dialogue formats do not currently provide. Practitioner voices from sectors actively managing AI deployment, including education, healthcare, and public administration, tend to be consulted after normative frameworks have already been shaped by technical and policy specialists. Reversing this sequence, by grounding agenda-setting in documented implementation experience before moving to principle development, may produce more durable outcomes. Format may include structured pre-dialogue submissions from practitioner communities with synthesis processes that preserve specificity rather than aggregating responses into broad themes. Plenary sessions benefit from being organised around concrete governance problems with documented evidence rather than abstract thematic areas. Working groups should include explicit representation from regulatory bodies, civil society organisations, and frontline practitioners alongside member state delegations and industry representatives.

Practitioner communities in education, health, and social services are substantially underrepresented, despite being responsible for implementing AI governance in contexts that directly affect large populations. Their exclusion means that governance frameworks tend to be evaluated against technical and economic criteria rather than against the conditions of actual deployment. Students, patients, and others who are subject to AI-assisted decisions are rarely present as anything other than the subject of discussion. Structured mechanisms for incorporating affected community perspectives, including through civil society proxies with genuine accountability to those communities, would address part of this gap. Indigenous communities and linguistic minorities are also underrepresented, despite facing disproportionate risks from AI systems trained on non-representative data.

Problem-based working sessions organised around documented governance failures, rather than thematic panels, tend to generate more substantive engagement. Participants are more likely to contribute meaningfully when discussion is grounded in specific cases with available evidence rather than in abstract principles that different stakeholders interpret differently. This format also makes it easier to identify where genuine disagreement exists and where apparent disagreement reflects differences in terminology rather than substance. Structured peer exchange between jurisdictions at comparable stages of governance development may produce more transferable learning than exchanges between highly asymmetric partners. When the gap in regulatory capacity and technical infrastructure between participants is very large, dialogue tends to default to knowledge transfer in one direction, which limits the quality of insight available to all parties and reinforces existing hierarchies in governance knowledge production. Asynchronous contribution mechanisms, including written submissions that receive substantive responses rather than symbolic acknowledgment, would allow participation from practitioners who cannot attend in person and whose time constraints currently exclude them from international dialogue processes. This is particularly relevant for frontline practitioners in education, health, and public administration, whose implementation experience is directly relevant to governance design but who rarely have institutional support for international travel or extended participation in formal dialogue processes. Deliberative formats that separate evidence review from position-taking may also reduce the degree to which submissions function as advocacy rather than as genuine contributions to shared understanding. Requiring participants to engage with evidence submitted by others before advancing their own positions is a relatively low-cost structural change that tends to improve the quality of deliberation in multi-stakeholder forums. Pilot testing engagement formats with diverse practitioner groups before the Dialogue convenes would allow organisers to identify which approaches generate the most substantive exchange across different participant profiles.​​​​​​​​​​​​​​​​

Several existing frameworks illustrate what effective AI governance looks like in practice and where implementation gaps remain. UNESCO's Recommendation on the Ethics of Artificial Intelligence, adopted in 2021 and applicable to all 194 member states, establishes human rights and dignity as the cornerstone of AI governance, with particular emphasis on transparency, fairness, and human oversight. What makes the Recommendation especially applicable are its Policy Action Areas, which allow policymakers to translate core values into action across data governance, education and research, and health and social wellbeing. This principle-to-action structure represents a model worth replicating in sector-specific governance development. The OECD AI Principles, initially adopted in 2019 and updated in May 2024, guide AI actors in developing trustworthy AI and provide policymakers with recommendations for effective AI policies, with countries using them to shape policies and create AI risk frameworks, building a foundation for global interoperability between jurisdictions. The EU AI Act, which entered into force in August 2024, sets out risk-based rules for AI developers and deployers, categorising risks into four levels and establishing compliance obligations that scale with potential harm. This risk-proportionate approach has influenced governance development beyond the EU, including in Australia. Australia's Voluntary AI Safety Standard, launched in September 2024, provides ten guardrails for organisations to manage AI risks, covering transparency, accountability, risk management, and supply-chain considerations. Experience in Australian higher education suggests that voluntary standards of this kind are adopted unevenly, and that their effectiveness depends on the availability of sector-specific implementation guidance rather than on the quality of the principles themselves. Across these frameworks, a consistent pattern emerges. Governance instruments that translate normative commitments into operationally specific guidance for defined deployment contexts tend to achieve higher uptake than general principles applied downward across sectors. The Dialogue may add value by supporting the development of this translation layer rather than producing further high-level frameworks that practitioners lack the infrastructure to implement.