RMIT School of Law

A successful first Global Dialogue on AI Governance would deliver clear, actionable, and inclusive outcomes, rather than broad statements of intent. First, it should establish baseline global principles on transparency, accountability, safety testing, and human oversight. While not necessarily binding, these standards must be specific enough to guide national regulation and industry practice. Second, success would involve practical coordination mechanisms. This includes agreement on shared terminology, information-sharing channels between governments (such as on AI incidents and risks), and early steps toward interoperable regulatory frameworks to avoid fragmentation. Third, the Dialogue should secure concrete commitments from industry. Major AI developers should agree to measurable actions, such as publishing risk assessments, conducting red-teaming, and adhering to audit standards, ensuring governance is not solely state-driven. Fourth, meaningful Global South inclusion is essential. Success requires not just representation, but real influence through capacity-building initiatives, funding commitments, and recognition of diverse economic and social impacts of AI systems. Fifth, the Dialogue should produce a forward-looking roadmap, including timelines for future meetings, dedicated working groups on key issues (such as frontier AI, data governance, and labour impacts), and mechanisms to track progress. Finally, success would be reflected in trust-building. If participants leave with greater confidence in one another's intentions and a shared understanding of risks, the Dialogue will have laid a strong foundation. In sum, the Dialogue succeeds if it moves from abstract concern to coordinated, credible, and inclusive action.

• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Interoperability of governance approaches
• Transparency, accountability, and human oversight
• Safe, secure and trustworthy AI

Prioritising transparency, accountability, and human oversight is essential because they are the mechanisms that make all other AI governance goals credible and enforceable. First, transparency underpins trust and effective regulation. Without visibility into how AI systems are designed, trained, and deployed, neither regulators nor affected individuals can assess risks, detect bias, or evaluate compliance. Transparency also enables interoperability across jurisdictions by creating shared expectations about documentation, reporting, and explainability. Second, accountability ensures that responsibility for AI harms is identifiable and enforceable. As AI systems become more complex and distributed across actors (developers, deployers, users), clear accountability frameworks prevent responsibility gaps. This is critical for maintaining public confidence and for providing meaningful remedies when harm occurs, particularly in high-stakes contexts such as housing, employment, or legal decision-making. Third, human oversight acts as a safeguard against over-reliance on automated systems. It ensures that critical decisions remain subject to human judgment, especially where ethical considerations, contextual nuance, or rights-based concerns arise. Human oversight is also key to managing unforeseen risks, system failures, or emergent behaviours that cannot be fully anticipated at the design stage. Together, these principles are enabling conditions for your four priorities. They operationalise "safe, secure and trustworthy AI" by making safety claims verifiable; they support capacity-building by establishing clear standards that jurisdictions can adopt; they help address the social, economic, and ethical implications of AI by embedding fairness and recourse; and they facilitate interoperability by aligning governance expectations across borders. In short, without transparency, accountability, and human oversight, AI governance risks becoming aspirational rather than effective.

Yes, while these priorities are comprehensive, several cross-cutting and emerging issues sit across them and warrant explicit attention. First, compute governance and concentration of power is increasingly central. The development of advanced AI systems depends on access to large-scale computational resources, which are concentrated among a small number of firms and states. This raises competition, security, and equity concerns that are not fully captured by capacity-building or interoperability alone. Second, data governance and provenance remains an unresolved issue. Questions around data ownership, consent, quality, and cross-border data flows cut across safety, ethics, and interoperability. Emerging challenges, such as the use of synthetic data and copyrighted materials in training, require clearer global norms. Third, evaluation and measurement standards are a critical gap. There is no widely accepted, harmonised framework for assessing AI risks, capabilities, or societal impact. Without shared benchmarks, claims about "safe" or "trustworthy" AI remain difficult to verify, undermining both accountability and interoperability. Fourth, labour market disruption and economic concentration deserves distinct attention. While social and economic impacts are included in your themes, the scale and speed of potential workforce transformation, alongside the concentration of economic gains, raise structural policy questions around redistribution, reskilling, and social safety nets. Fifth, misinformation and information integrity is an urgent, rapidly evolving issue. Generative AI systems can amplify disinformation at scale, affecting elections, public discourse, and social cohesion, which may require targeted governance responses beyond general ethical considerations. Finally, environmental and energy impacts of AI are often overlooked. The growing computational demands of advanced models have significant carbon and resource implications, linking AI governance to broader sustainability agendas. These issues are "cross-cutting" because they intersect with, and in some cases condition, the success of your four priorities. Explicitly recognising them would strengthen the overall coherence and forward-looking nature of the framework.

In the Australian context, governance gaps across these thematic areas are already shaping both risks and opportunities. A key challenge is the fragmented and evolving regulatory landscape. While Australia has advanced principles-based frameworks, such as the Australian Government's AI Ethics Principles, there is no comprehensive, enforceable regime governing high-risk AI. This creates uncertainty for businesses and uneven protections for individuals, particularly in sectors like employment, housing, and financial services. Second, capacity constraints remain significant. Although Australia has strong research institutions, there is limited sovereign capability in developing frontier AI systems. This increases reliance on foreign technologies and raises concerns around data sovereignty, security, and strategic dependence, particularly in sensitive sectors. Third, the social and economic impacts are becoming more visible. AI-driven automation is reshaping white-collar professions, including legal and administrative work, creating both productivity gains and workforce disruption. At the same time, risks of bias and opacity in automated decision-making systems may exacerbate existing inequalities if not carefully governed. Fourth, interoperability challenges are acute. Australian organisations must navigate a patchwork of international regimes (e.g., the EU and US approaches), increasing compliance burdens and complicating cross-border data flows and AI deployment. However, these gaps also create opportunities. Australia is well-positioned to act as a norm entrepreneur, aligning with trusted international partners while shaping practical, risk-based governance models. There is also an opportunity to invest in AI capacity-building, particularly in public sector capability and workforce reskilling, to ensure broad-based benefits. Finally, strengthening transparent and accountable AI practices could enhance public trust and support innovation. If addressed strategically, current governance gaps could become a catalyst for building a more resilient, competitive, and inclusive AI ecosystem.

The AI Dialogue can play a pivotal role as a coordination and bridge-building forum, turning shared concerns into practical international cooperation. First, it can promote norm convergence by aligning baseline standards on safety, transparency, and accountability. Even without binding rules, soft-law tools, such as joint statements or voluntary codes, can meaningfully shape state and industry behaviour. Second, it can advance regulatory interoperability. Through technical working groups, participants can develop common definitions, risk classifications, and reporting standards, reducing fragmentation and easing cross-border compliance. Third, the Dialogue can strengthen information-sharing and collective risk management by creating channels to share insights on AI incidents, emerging risks, and best practices, particularly for high-impact systems. Fourth, it can support inclusive capacity-building by coordinating funding, technical assistance, and knowledge exchange, ensuring developing countries can meaningfully participate in shaping AI governance. Fifth, it provides a platform for multi-stakeholder engagement, bringing together governments, industry, academia, and civil society to ensure governance is both technically informed and socially legitimate. Finally, the Dialogue can maintain strategic continuity through a forward-looking roadmap, including ongoing working groups and mechanisms to track progress. Overall, its value lies in moving from fragmented national approaches to coordinated, interoperable, and inclusive global governance, while remaining flexible in the face of rapid technological change.

The AI Dialogue should build on existing multilateral and multi-stakeholder initiatives to avoid duplication and accelerate convergence. Key frameworks include the OECD AI Principles, which provide widely endorsed normative guidance; the G7 Hiroshima AI Process, focused on generative AI governance; and the United Nations' emerging work on global digital cooperation. Regional regimes, such as the EU's risk-based approach, and technical bodies like ISO and IEEE also play a critical role in shaping standards and operational practices. The added value of the AI Dialogue lies in connecting these fragmented efforts. It can act as a coordination layer, aligning principles with technical standards and policy implementation. Unlike existing forums, it can integrate policy, technical, and geopolitical perspectives in one setting, while maintaining flexibility through non-binding cooperation. It can also enhance inclusivity, ensuring that developing countries have a stronger voice than in smaller groupings like the G7. Finally, by focusing on interoperability and practical outcomes, such as shared definitions, reporting mechanisms, and risk assessment frameworks, the Dialogue can translate existing principles into more consistent global practice.

Different stakeholders should contribute in ways that reflect their expertise while ensuring balanced and inclusive participation. Governments should provide policy leadership, align national approaches, and commit to information-sharing on AI risks and incidents. Industry should contribute technical expertise and make measurable commitments, such as transparency reporting, safety testing, and auditing practices. Academia can offer independent research, evaluation methodologies, and foresight on emerging risks. Civil society plays a critical role in representing affected communities, highlighting rights-based concerns, and ensuring accountability. International organisations can help coordinate standards, aggregate knowledge, and support capacity-building, particularly for developing countries. In terms of format and structure, the Dialogue should be designed for both flexibility and continuity. A multi-track model would be effective: A high-level plenary to set strategic priorities and maintain political momentum Thematic working groups (e.g., safety, capacity-building, interoperability) to develop technical outputs and recommendations Multi-stakeholder roundtables to ensure diverse perspectives are integrated into decision-making The Dialogue should also adopt a hybrid and iterative structure. Regular meetings (e.g., annually at the ministerial level, with ongoing working group engagement) would allow for sustained progress. Outputs should be practical and trackable, such as voluntary commitments, model guidelines, and shared standards, supported by light-touch monitoring mechanisms. To ensure inclusivity, the Dialogue should incorporate capacity-building support (e.g., funding, training, and technical assistance) to enable meaningful participation from the Global South. Overall, an effective AI Dialogue would combine broad participation, technical depth, and continuous engagement, ensuring that discussion translates into coordinated and credible action.

Several voices remain underrepresented in global AI governance discussions, limiting both legitimacy and effectiveness. First, Global South countries often have limited participation. This is due to resource constraints, technical capacity gaps, and a historical focus on governance frameworks developed in high-income states. Without their input, global standards risk being misaligned with diverse social, economic, and cultural contexts. Inclusion could be strengthened through targeted capacity-building, technical assistance, and funded participation programs that enable meaningful engagement in working groups and plenaries. Second, marginalised communities within countries, including Indigenous peoples, low-income groups, and minority linguistic or cultural groups, are rarely consulted. These populations experience disproportionate impacts from AI systems, such as bias in automated decision-making, but lack mechanisms to voice their concerns. Participatory consultations, citizen assemblies, and multi-lingual outreach can amplify these perspectives. Third, smaller enterprises and start-ups often lack access to international policy fora, meaning governance rules may favour large tech companies. Structured engagement channels, advisory councils, or innovation labs can help incorporate their technical and operational perspectives. Fourth, non-technical disciplines, such as social scientists, ethicists, and humanities scholars, are underrepresented, yet they offer crucial insight into societal impacts, fairness, and cultural context. Including them in thematic working groups and scenario exercises would ensure a more holistic approach. Finally, youth and future-focused groups are rarely given a voice, despite being most affected by long-term AI impacts. Mechanisms such as youth advisory panels or global hackathons can embed their perspectives in policy discussions. Inclusion of these voices requires deliberate, structured, and resourced approaches, from funded participation and multi-stakeholder workshops to citizen engagement and capacity-building initiatives. By doing so, the AI Dialogue can produce governance frameworks that are legitimate, equitable, and globally applicable, reflecting the full range of human and societal interests.

To foster meaningful and dynamic engagement, the AI Dialogue should combine interactive, multi-stakeholder, and technology-enabled formats rather than relying solely on traditional plenaries. Scenario-based workshops can simulate real-world AI risks and governance challenges, allowing participants to test policies and technical safeguards in a controlled environment. These exercises help stakeholders understand trade-offs, build shared understanding, and surface practical solutions. Multi-stakeholder roundtables bring together governments, industry, academia, and civil society to deliberate on specific thematic issues, such as safety, interoperability, or social impacts. Structured dialogue with facilitators can ensure all voices, including those from the Global South, are heard. Hackathons and collaborative sprints focused on technical standards, auditing frameworks, or data-sharing protocols can generate concrete outputs quickly. These formats encourage hands-on problem solving and foster innovation through collaboration between policymakers and technologists. Digital engagement platforms can broaden participation, allowing asynchronous input from a wider set of stakeholders and real-time feedback during sessions. Interactive polling, AI-assisted discussion summarisation, and collaborative document editing can make virtual participation more dynamic. "Fishbowl" discussions, where a small group debates an issue in view of a larger audience, can surface differing perspectives while encouraging broader observation and reflection. Finally, integrating continuous working groups and follow-up labs ensures that insights from the Dialogue are acted upon and iteratively refined. Outputs from these labs can feed back into future Dialogue sessions, creating a cycle of learning and implementation. By combining experiential exercises, technology-enabled tools, and structured deliberation, these formats can promote active participation, cross-sector collaboration, and practical problem-solving, making the AI Dialogue not only a forum for discussion but a catalyst for actionable global AI governance.

Several existing policies and practices offer concrete, scalable approaches to effective AI governance. First, risk-based regulatory frameworks-most prominently the EU's AI Act-demonstrate how to tier obligations based on risk level, with stricter requirements for high-risk systems (e.g., in employment or critical infrastructure). This approach balances innovation with safeguards and provides a clear compliance pathway for industry. Second, the OECD AI Principles offer widely adopted normative guidance, particularly on transparency, accountability, and human-centred values. Their influence shows the value of non-binding but globally endorsed standards. Third, algorithmic impact assessments (AIAs), used in jurisdictions like Canada, provide a practical tool for identifying and mitigating risks before deployment. These assessments can be standardised and adapted across sectors, strengthening accountability. Fourth, model evaluation and red-teaming practices adopted by leading AI developers illustrate how technical safeguards can be embedded into development cycles. Structured stress-testing of systems helps identify vulnerabilities, including bias, misuse risks, and safety failures. Fifth, multi-stakeholder governance platforms, such as the Global Partnership on AI, demonstrate the value of combining policy, technical expertise, and civil society input to produce actionable recommendations and share best practices. Sixth, transparency mechanisms, including model cards and system documentation, provide accessible information about how AI systems function, their limitations, and appropriate use cases, supporting both regulatory oversight and public trust. Finally, regulatory sandboxes offer a flexible approach, allowing innovators to test AI systems under regulatory supervision while enabling policymakers to better understand emerging technologies. Together, these examples highlight that effective AI governance combines risk-based regulation, practical tools, technical safeguards, and collaborative platforms, translating high-level principles into operational and enforceable outcomes.