The Libyan Academy

A successful inaugural Global Dialogue on AI Governance would hinge on three key outcomes that move the conversation beyond mere rhetoric toward tangible progress. First, success would be establishing a shared minimum common ground. The Dialogue should yield a formal, though potentially concise, consensus document acknowledging core principles. This must go beyond vague aspirations to include a shared definition of "high-risk" AI applications and a mutual commitment to existing international human rights frameworks as the irreducible baseline for all subsequent governance efforts. Simply put, we need to agree on what we are talking about and the fundamental values we are protecting. Second, the Dialogue must produce a concrete roadmap for interoperability. A fragmented landscape of competing national regulations would be a disaster for innovation and a boon for regulatory arbitrage. Success means initiating a structured, time-bound process to map different emerging regulatory approaches (like the EU's risk-based model and the US's sectoral approach) to identify where they can be made interoperable. The goal is not one global rulebook, but a "plug-and-play" architecture that allows different systems to work together seamlessly, preventing a "balkanization" of the internet and AI supply chains. Finally, and most critically, the Dialogue must bridge the global majority divide. It cannot be another forum where the Global North dictates terms to the Global South. True success means securing concrete commitments to capacity-building: funding for AI safety institutes in developing nations, technology transfer for auditing tools, and a dedicated mechanism to ensure that the voices and unique concerns of the Global South—from labor displacement to data sovereignty—are not just heard but are structurally embedded in the ongoing governance process. Without this, any outcome will be inherently fragile and incomplete.

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

My selection of these four priorities reflects a strategic focus on both foundational enablers and critical safeguards necessary for responsible AI governance. Safe, secure and trustworthy AI is the non-negotiable prerequisite for all else. Without confidence in the reliability of AI systems, neither adoption nor governance can proceed effectively. This priority addresses the immediate technical vulnerabilities that could undermine public trust and cause real-world harm, making it the essential foundation upon which all other governance efforts must build. AI capacity-building addresses the profound global asymmetry in AI capabilities. The current concentration of expertise risks creating a two-tier world of AI "haves" and "have-nots." Urgent action here is about democratic legitimacy-ensuring that all nations can participate meaningfully in shaping AI's future rather than merely accepting standards set elsewhere. It transforms governance from a passive exercise into an active, inclusive endeavor. Interoperability of governance approaches responds to the practical reality of a connected world. Without deliberate coordination, divergent regulatory requirements will create compliance burdens that stifle innovation while doing little to enhance safety. This priority seeks to prevent fragmentation, enabling responsible AI development to flourish across borders while maintaining robust protections. Transparency, accountability, and human oversight operationalize ethical commitments. These mechanisms transform abstract principles into enforceable practice. They ensure that when AI fails-as it inevitably will-there are clear lines of responsibility, meaningful recourse for affected individuals, and human judgment applied to consequential decisions. This priority safeguards human agency in an increasingly automated world.

While the thematic areas identified in Resolution 79/325 are comprehensive, several cross-cutting and emerging issues warrant explicit attention as they permeate all governance discussions. Energy consumption and environmental impact represents a critical gap. The carbon footprint of large-scale AI training and inference is substantial and growing, yet environmental sustainability does not appear as a distinct consideration within the listed themes. This is both a cross-cutting concern-relevant to capacity-building (developing countries cannot afford energy-intensive infrastructure), interoperability (different energy standards affect deployment), and safe AI (reliability under energy constraints)-and an emerging crisis requiring urgent attention. Future dialogues must address how to govern AI's environmental costs alongside its benefits. Labor displacement and just transition merits more prominent treatment. While "social and economic implications" gestures toward this, the scale of potential workforce disruption demands focused consideration. This includes not only job losses but also questions about AI-driven changes to work quality, surveillance in workplaces, and mechanisms for ensuring that productivity gains are broadly shared rather than concentrated. This connects directly to human rights, transparency (algorithmic management), and capacity-building for workforce adaptation. Geopolitical stability and dual-use governance sits uncomfortably between the identified themes. AI's military applications, potential for autonomous weapons, and implications for strategic stability require dedicated governance attention. This includes export controls, proliferation risks, and the intersection of civilian and military AI development. The civil society voice is often excluded from these discussions, yet the implications for global security affect everyone. Finally, impacts on democratic processes and information integrity deserves explicit recognition. AI-generated disinformation, synthetic media, and algorithmic amplification of harmful content threaten democratic deliberation worldwide. This spans human rights (access to reliable information), transparency (content provenance), and safety (societal resilience).

As an Academic organization operating globally, we observe governance gaps across our four priority areas creating significant challenges while also presenting opportunities for constructive engagement. Safe, secure and trustworthy AI: The absence of binding safety standards means we increasingly document AI harms—from discriminatory lending algorithms to unsafe autonomous systems—without adequate recourse mechanisms. Regions with stronger consumer protection frameworks are moving faster to address these gaps, creating uneven protection for affected communities. The opportunity lies in building evidence-based advocacy for harmonized safety standards that protect all users regardless of geography. AI capacity-building: The governance gap here is stark—capacity-building remains underfunded and ad hoc while AI capabilities concentrate in a few jurisdictions. In our sector, this means civil society organizations in the Global South lack the technical expertise to meaningfully engage with AI governance discussions affecting their communities. The opportunity is to develop open educational resources, facilitate South-South knowledge exchange, and advocate for dedicated capacity-building funding as a governance priority rather than an afterthought. Interoperability of governance approaches: The current fragmentation creates compliance burdens that disproportionately affect smaller actors, including civil society organizations developing AI tools for social good. We navigate conflicting requirements across jurisdictions while larger technology companies resource compliance teams. The opportunity is to champion interoperability frameworks that reduce friction for responsible innovation while maintaining robust protections. Transparency, accountability, and human oversight: Meaningful transparency remains elusive, with companies providing insufficient information about training data, model behavior, and deployment contexts. This gap directly impedes our ability to audit systems for bias or harm. Without accountability mechanisms, affected communities lack recourse when AI systems cause harm. The opportunity is to develop and promote transparency standards that enable independent oversight while protecting legitimate business interests.

The AI Dialogue can serve as a critical catalyst for international cooperation, but its success depends on function, not form. It must move beyond being merely a convening space to become a genuine bridge-builder between fragmented governance efforts. First, the Dialogue can establish shared normative foundations. By providing a universally endorsed platform under UN auspices, it can forge consensus on irreducible principles—respect for human rights, environmental sustainability, and human dignity—that must underpin all AI governance. This normative baseline creates a common language and reference point for all subsequent cooperation. Second, it can enable regulatory interoperability through structured dialogue. Rather than pursuing a single treaty (a distant prospect), the Dialogue should facilitate mutual understanding between different regulatory approaches—the EU's risk-based framework, the US's sectoral model, China's state-directed approach, and Global South perspectives. This involves mapping convergences, identifying friction points, and developing guidance for cross-system compatibility. The Dialogue's legitimacy can encourage jurisdictions to adapt frameworks for interoperability voluntarily. Third, it can mobilize capacity-building at scale. The Dialogue can aggregate demand from developing countries, connect them with technical assistance providers, and mobilize resources from donor nations and philanthropic organizations. It can establish peer-learning networks and facilitate technology transfer on favorable terms. This transforms capacity-building from rhetoric into measurable outcomes. Fourth, it can provide an early warning and rapid response mechanism for emerging risks. Through multistakeholder participation, the Dialogue can identify novel AI applications with cross-border implications—synthetic media threatening elections, autonomous systems in conflict zones, labor displacement at scale—and facilitate coordinated international responses before harms crystallize. Finally, the Dialogue can amplify marginalized voices. By ensuring meaningful participation from civil society, the Global South, and affected communities, it can counterbalance the dominance of powerful states and corporations in AI governance discussions, producing more legitimate and durable outcomes.

The AI Dialogue should strategically connect with existing initiatives rather than duplicating them, focusing on its unique value as a universal, intergovernmental platform with legitimacy that no other forum possesses. Key initiatives to build upon include the OECD AI Principles and its AI Policy Observatory, which provide substantive expertise and policy tracking. The Global Partnership on AI (GPAI) offers multistakeholder technical work and implementation guidance. UNESCO's Recommendation on the Ethics of AI provides normative grounding with 193 member states' endorsement. Regional bodies like the African Union and Council of Europe (developing the first binding AI treaty) offer regionally-specific insights. Technical standard-setting bodies like ISO/IEC and IEEE provide essential technical foundations. Multistakeholder initiatives such as the Partnership on AI and AI for Good Global Summit bring civil society and technical communities into governance discussions. The Dialogue's added value lies in four distinct contributions: Universality and legitimacy: As a UN process with universal membership, the Dialogue can validate and synthesize work from other forums, giving it political weight that regional or technical bodies cannot achieve alone. Connecting fragmented efforts: Currently, AI governance resembles an archipelago of disconnected initiatives. The Dialogue can serve as a hub, facilitating coherence between technical standards, regional regulations, and normative frameworks. Amplifying Global South voices: Existing forums are disproportionately Northern. The Dialogue can intentionally elevate perspectives from developing countries, ensuring governance reflects global rather than parochial interests. Political accountability: Technical discussions in other forums require political endorsement to translate into action. The Dialogue can provide this bridge, converting expert consensus into political commitment. Early warning and rapid response: No existing mechanism adequately addresses emerging cross-border AI risks. The Dialogue can fill this gap through coordinated monitoring and response capabilities.

Effective multistakeholder participation requires intentional design that moves beyond tokenism to genuine influence. Different stakeholders bring distinct contributions that the Dialogue's structure must actively enable. Member States provide political legitimacy and implementation authority. Their role is to translate consensus into binding commitments, allocate resources for capacity-building, and ensure domestic alignment with international frameworks. They should lead intergovernmental negotiations while creating space for other voices. Civil Society offers ground-truth perspectives on AI's impacts, particularly for marginalized communities. We document harms, advocate for rights-based approaches, and hold powerful actors accountable. Our contribution requires safe participation channels, protection from reprisals, and meaningful inclusion in both formal sessions and informal negotiations. Private Sector contributes technical expertise, implementation experience, and innovation insights. Responsible companies can demonstrate best practices, identify feasible standards, and commit to voluntary safeguards. Their participation must be structured to prevent regulatory capture and ensure competition concerns don't silence smaller actors. Technical Communities (researchers, engineers, standards bodies) provide essential expertise on what is technically possible, measurable, and verifiable. They can translate policy goals into technical requirements and identify emerging capabilities before they reach deployment. International Organizations offer convening power, secretariat support, and linkages to existing frameworks. They can provide continuity between Dialogue sessions and connect AI governance to related domains like trade, labor, and human rights. For structure, I recommend: Hybrid participation model enabling both in-person and virtual engagement to maximize inclusivity, with interpretation and accessibility provisions. Preparatory process including regional consultations to surface diverse perspectives before formal negotiations, with synthesized inputs feeding directly into deliberations. Multistakeholder advisory body providing ongoing expertise and ensuring non-state voices shape the agenda, with balanced geographic and sectoral representation. Working groups on priority themes enabling deep technical work between plenary sessions, with multistakeholder composition and clear mandates. Transparent documentation including webcasts, timely posting of submissions, and accessible summaries in multiple languages.

Global AI governance discussions suffer from systematic exclusions that undermine both legitimacy and effectiveness. Addressing these gaps is essential for producing outcomes that serve humanity collectively rather than particular interests. The Global South remains dramatically underrepresented. Most governance forums convene in the North, operate in English, and frame problems through Northern lenses. Yet AI's impacts—labor displacement, data extraction, cultural homogenization—often hit developing countries hardest. Inclusion requires dedicated regional consultations, funding for participation, translation and interpretation, and governance structures with guaranteed majority-world representation. Affected communities—those directly experiencing AI harms—are rarely at the table. Workers subjected to algorithmic management, communities targeted by predictive policing, patients subjected to biased diagnostic tools—their lived expertise is essential for understanding what "harm" means in practice. Inclusion requires partnerships with grassroots organizations, community hearing formats, and protections against retaliation for speaking truth to power. Informal workers and labor organizers face unique AI-driven challenges including algorithmic surveillance, platform-mediated precarity, and erosion of bargaining power. Traditional labor voices are often absent from technology governance discussions. Inclusion requires dedicated labor tracks, partnerships with global union federations, and explicit consideration of worker perspectives in all thematic areas. Indigenous peoples bring distinctive perspectives on data sovereignty, collective rights, and relationships with knowledge that challenge Western individualistic frameworks. Their traditional knowledge and governance models offer alternatives to dominant approaches. Inclusion requires adherence to Indigenous data sovereignty principles, free prior and informed consent protocols, and dedicated engagement mechanisms. Women and gender-diverse individuals face compounded AI harms including reinforcement of stereotypes, exclusion from development, and disproportionate surveillance. Yet gender analysis remains marginal in most governance discussions. Inclusion requires gender-balanced representation, dedicated gender analysis of all proposals, and safe participation mechanisms free from harassment. Small and medium enterprises and civil society organizations from all regions lack the resources of large corporations and states to track and influence governance processes. Inclusion requires capacity-building support, simplified engagement mechanisms, and protection from resource asymmetries that silence critical voices.

To move beyond staid diplomatic rituals toward genuine exchange, the AI Dialogue should embrace innovative formats that surface diverse perspectives and build trust across difference. World café conversations arrange participants in small, rotating groups for facilitated discussion on specific questions. This format democratizes participation—every voice can be heard, hierarchies dissolve, and cross-pollination of ideas occurs naturally. For AI governance, this could surface insights from participants who would never speak in plenary. Scenario-based foresight exercises immerse participants in plausible future scenarios—a major AI failure, a breakthrough in general intelligence, a regulatory race to the bottom. Teams composed across stakeholder groups work through response strategies, revealing assumptions, values, and priorities that abstract principles discussions miss. The shared experience of navigating uncertainty builds relationships and identifies common ground. Design sprints adapted from product development bring diverse teams together to prototype governance solutions—a transparency reporting template, a capacity-building curriculum, an interoperability framework. The hands-on, time-bound format focuses energy on tangible outputs and demonstrates what multistakeholder collaboration can achieve. Community hearing formats invert traditional power dynamics by placing affected communities at the center. Rather than communities testifying to experts, experts listen as communities share experiences of AI impacts in their own words, using their own frames. This models the epistemic humility essential for legitimate governance. Ignite sessions feature rapid-fire presentations (20 slides, 15 seconds each) from diverse voices, conveying passion and expertise efficiently while exposing participants to perspectives they might otherwise miss. Peer assist methodology, borrowed from knowledge management, invites participants facing similar challenges to share experiences and solutions in structured, solution-oriented sessions. A country developing capacity-building programs, for example, receives insights from others further along the path. Art and storytelling spaces recognize that governance engages values and emotions, not just technical analysis. Exhibitions, performances, and storytelling circles can communicate AI's human impacts in ways policy papers cannot, fostering empathy and shared purpose.

Drawing from diverse jurisdictions and sectors, several concrete policies, practices, and platforms demonstrate effective AI governance in action. These examples span regulatory frameworks, institutional mechanisms, operational tools, and capacity-building approaches. Regulatory frameworks with contextual adaptation: Italy's national AI law complements the EU AI Act by addressing sector-specific priorities . It prohibits AI for selecting access to healthcare services, requires patient information about AI use, and mandates that AI remains merely supportive to clinical decisions with doctors retaining ultimate responsibility. In employment, it requires employers to inform workers about AI use and establishes a National Observatory to monitor labor impacts. The law also explicitly extends copyright protection to AI-assisted works, provided human intellectual work predominates . Municipal governance innovation: New York City's GUARD Act establishes the first municipal AI oversight office with enforceable standards rather than voluntary guidelines . It creates an independent Office of Algorithmic Data Accountability to audit agency AI tools, investigate public complaints, and publish a public directory of all reviewed AI systems. Mandatory citywide standards require fairness testing, privacy protections, and independent evaluation before deployment . This demonstrates that governance can be implemented at local levels with real enforcement teeth. Enterprise governance practices: Microsoft's sensitive use case framework provides a practical model for operationalizing ethical principles . It categorizes sensitive uses as those involving denial of consequential services, risk of harm, or infringement on human rights. When a law enforcement agency requested facial recognition, Microsoft's multi-stage review process-involving Responsible AI Champs, the Aether Committee, and senior leadership-led to approving only appropriately safeguarded use cases while declining support for uncontrolled identification scenarios . Regulatory sandboxes for innovation support: Denmark's Regulatory Sandbox for AI helps businesses and public authorities develop AI projects by offering expert advice on data protection and AI rules . It reduces legal uncertainty, speeds responsible market entry, and will expand to include AI Act guidance. This represents proactive, future-oriented governance that balances innovation with legal and ethical responsibilities . Capacity-building through collaborative platforms: Thailand's AI Governance Clinic bridges the gap between abstract principles and practical application . It operates through a fellowship program connecting local experts across sectors, develops actionable implementation toolkits including "AI Governance Guidelines for Executives," and offers specialized executive education for healthcare administrators. The clinic emphasizes interdisciplinary "clinical" components where participants explore real-world implementation challenges across varied contexts . Technical tools for operationalizing governance: The OECD AI Catalogue of Tools & Metrics features numerous platforms for implementing trustworthy AI . Vectice provides regulatory MLOps for streamlining documentation and audit readiness. FairNow centralizes AI risk management at scale. Enzai offers EU AI Act compliance frameworks breaking regulations into actionable steps. The Data Carbon Ladder assesses and measures data CO2 footprints to support environmentally sustainable AI . Technical governance infrastructure: Governor is an open-source AI Governance Control Tower that sits between AI agents and tools, enforcing policy decisions in real-time while capturing telemetry for observability and compliance . It provides an "AI Action Firewall" with policy engine, risk classification, and approval workflows, demonstrating how technical architecture can embed governance directly into AI operations . These examples collectively illustrate that effectiv