Solort

For the first Global Dialogue on AI Governance to be considered a genuine success, the outcome must move beyond "shared principles" and into the realm of interoperable frameworks. In my view, three specific benchmarks would define that success: 1. Interoperability We don't need a single global law, that's unrealistic given differing legal traditions. Success looks like a commitment to cross-border compatibility. If the EU's risk-based approach, the US's executive orders, and China's targeted regulations can "talk" to one another, we prevent a fractured digital landscape where innovation hits a wall at every border. 2. Agreed Safety Triggers Success requires a clear, non-negotiable consensus on existential risks. This means establishing a Global Safety Trigger: a unified agreement on what constitutes "unacceptable risk" (e.g., autonomous bioweapon synthesis or non-human-intervenable cyber warfare) and a shared protocol for pausing or auditing models that cross those thresholds. 3. Bridging the Digital Divide The dialogue is a failure if it remains a "G7-plus-one" conversation. A successful outcome must include a Resource Equity Roadmap, ensuring the Global South isn't just a consumer of AI or a source of raw data, but a participant in governance with access to the compute and "sovereign AI" capabilities needed to solve local challenges.

• Safe, secure and trustworthy AI
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Open-source software, open data and open AI models
• Interoperability of governance approaches

1. Safe, Secure, and Trustworthy AI This is the foundational "license to operate." Without a global consensus on safety standards and risk mitigation, public trust collapses. The Dialogue aims to move from vague ethics to concrete technical standards that prevent AI from being used for malicious purposes or causing unintended systemic harm. 2. Multi-Dimensional Implications AI isn't just a technical challenge; it's a cultural and economic one. By prioritizing the social, ethical, and linguistic impacts, the Dialogue ensures that AI doesn't just reflect the values (or languages) of the "Big Tech" hubs, but respects global diversity and protects labour markets in the Global South. 3. Interoperability of Governance As shown in the diagram below, the current global landscape is a "patchwork" of different laws (like the EU AI Act vs. US Executive Orders). Interoperability is the goal of creating "bridges" between these systems so that companies and researchers can innovate globally without facing 193 different sets of conflicting rules. 4. Open-Source and Open Data This is the key to equity. By championing open-source software and open AI models, the Dialogue seeks to prevent a "digital monopoly." It allows developing nations to build their own "Sovereign AI" using shared global knowledge rather than remaining entirely dependent on proprietary, expensive black-box systems.

1. The Environmental "Double Materiality" While "technical implications" are listed, the sheer scale of AI's resource footprint is becoming a stand-alone crisis. The Issue: AI data centres are projected to triple their energy consumption by 2035. The Gap: Current themes focus on what AI does, but not what it costs the planet in terms of water stress (for cooling) and e-waste. A success for the Dialogue would be a global standard for Environmental Impact Reporting that goes beyond voluntary disclosures. 2. Information Integrity & "Synthetic Reality" The theme of "Human Rights" covers privacy, but it doesn't fully capture the systemic threat to shared reality. The Issue: In a year with global elections, AI-generated disinformation isn't just about "fakes"-it's about the "Liar's Dividend," where people stop believing real evidence because everything could be fake. The Gap: This requires a specific focus on Content Provenance (watermarking/metadata) that is interoperable across all platforms, ensuring users can verify the source of information. 3. Long-term Safety and "Recursive" Risks "Safe and Trustworthy AI" often focuses on current harms like bias. However, the technical community is increasingly worried about Advanced AI Safety. The Issue: Risks related to agentic AI-systems that can autonomously set goals, replicate, or bypass human-coded "red lines." The Gap: There is no explicit mention of Emergency Preparedness Protocols or "Kill Switches" for models that demonstrate self-improving capabilities that exceed human control.

Significant Challenges The "Agentic" Governance Gap: A major challenge in the UK is the mismatch between the rapid deployment of autonomous AI agents in sectors like finance and healthcare and the "pilot-stage" maturity of oversight frameworks. Many UK firms are struggling with "audit gaps" where AI makes micro-decisions too fast for traditional human-in-the-loop systems. Regulatory Fragmenting: While the UK maintains a "pro-innovation" sector-led approach, the EU AI Act's enforcement cycle in 2026 is creating a "Brussels Effect." UK companies operating globally face the burden of complying with stringent EU high-risk classifications, making interoperability a survival necessity rather than a luxury. Copyright Uncertainty: Despite the Data (Use and Access) Act 2025, the UK government's 2026 reports show continued "can-kicking" on AI and copyright reform, leaving creators and AI developers in a state of prolonged legal uncertainty. Significant Opportunities Sovereign AI Infrastructure: The UK's newly established Sovereign AI Unit and the expansion of supercomputing (like the DAWN system in Cambridge) offer a massive opportunity to build domestic models that are "safe by design," reducing dependence on external proprietary systems. The "Accountability Dividend": There is a growing sector in the UK for AI Assurance. Companies that successfully bridge the governance gap are finding a competitive advantage, using transparency as a "trust multiplier" to win international contracts. Open-Source Leadership: By prioritizing open AI models, the region is fostering "AI Factories" that allow SMEs to innovate without the prohibitive costs of closed-box technologies, directly supporting the "AI Continent Action Plan."

1. Universalizing the Conversation Before this Dialogue, AI governance was largely the domain of the "G7" or "OECD" clubs. The Dialogue provides the first universal platform where all 193 UN Member States-particularly from the Global South-have an equal seat at the table. By including 118 countries previously excluded from major frameworks, it ensures that global standards reflect diverse economic and cultural realities rather than just those of Silicon Valley or Brussels. 2. Operationalising Interoperability A central goal of the Dialogue is to move past conflicting regulations. It serves as a "clearinghouse" for: Best Practices: Sharing what works in national legislation to prevent every country from "reinventing the wheel." Technical Baselines: Establishing common disclosure and transparency requirements so that a "trustworthy" model in one jurisdiction is recognized as such in another, reducing barriers to global trade and innovation. 3. Merging Science with Policy The Dialogue acts as the political recipient of the Independent International Scientific Panel on AI. By providing a shared "evidence engine," the Dialogue ensures that international negotiations are based on objective, peer-reviewed data about AI risks and capabilities. This prevents "regulatory capture" by a few powerful tech firms and grounds global cooperation in scientific reality.

1. Existing Initiatives to Build Upon The G7 Hiroshima AI Process (HAIP): The Dialogue should adopt the HAIP's International Code of Conduct and its voluntary reporting framework as a baseline for corporate transparency, scaling it from 66 "Friends" to all 193 UN states. OECD AI Principles & GPAI: The Dialogue should utilize the OECD's robust AI Incident Database and its risk-classification frameworks to provide a shared technical language for the "Independent Scientific Panel." The AISI International Network: By connecting with national AI Safety Institutes (like those in the UK, US, and Japan), the Dialogue can facilitate "safety diplomacy," sharing red-teaming results and evaluation protocols with developing nations. ITU's "AI for Good": Since the Dialogue is hosted back-to-back with this summit, it should leverage the ITU's existing technical standards to ensure linguistic and cultural diversity in AI datasets. 2. The Unique "Added Value" of the UN Dialogue The Dialogue provides what no other forum currently offers: Universal Legitimacy. Neutralizing the "Brussels Effect": While the EU AI Act sets a high bar, the Dialogue offers a space to negotiate mutual recognition so that regional laws don't become accidental trade barriers. Closing the Inclusion Gap: It moves governance from a "G7-plus" conversation to one where the Global South isn't just a consumer, but a co-designer of rules. Sovereign Data Governance: It provides a platform to discuss Open Data and Open Models as a public good, preventing a permanent "compute-divide" between a few wealthy nations and the rest of the world.

1. Stakeholder Contributions Member States: Focus on "Interoperability." National regulators should share legislative "sandboxes" and templates to reduce global fragmentation. Private Sector: Move from principles to "Red Lines." Tech firms should provide technical APIs and data access for independent auditing by the Scientific Panel. Civil Society & Academia: Act as the "Human Rights Watchdog." They must contribute evidence on the social, cultural, and linguistic impacts of AI, ensuring local communities aren't side-lined by global standards. Technical Community: Provide the "Open-Source" backbone. They should lead the development of shared global benchmarks for model safety and transparency. 2. Recommended Format & Structure The Dialogue should follow a modular, three-tier structure to balance high-level diplomacy with technical depth: I. The Evidence Plenary: Presentation of the Scientific Panel's Annual Report. Establish a shared, objective "fact-base" before negotiations begin. II. Thematic Working Groups: Small-group "Deep Dives" (e.g., Open Source, Capacity Building).Produce actionable "Outcome Documents" (e.g., global compute-sharing frameworks). III. High-Level Segment: Ministerial-level negotiations in the margins of the AI for Good Summit. Convert technical consensus into political commitments and funding. Strategic Suggestion: The Dialogue should utilize a "Hybrid-Local" model, holding satellite consultations in regional hubs (e.g., Nairobi, Santiago, Bangkok) to ensure stakeholders in the Global South can participate without the financial burden of traveling to Geneva or New York.

1. The "Global Majority" (Non-G7 Civil Society) While the 2026 India AI Impact Summit highlighted the "Majority World," many grassroots organizations from Africa, Southeast Asia, and Latin America are still side-lined. The Gap: Participation is often limited by travel costs and visa barriers to Geneva/New York. Inclusion Strategy: Establish Regional Consultative Hubs and a dedicated UN fund to subvent the participation of SMEs and NGOs from developing nations. 2. Indigenous Communities As AI models increasingly scrape cultural data, Indigenous peoples face "data extraction" without consent. The Gap: AI governance often treats data as a commodity rather than a cultural asset. Inclusion Strategy: Formally adopt the CARE Principles (Collective Benefit, Authority to Control, Responsibility, Ethics) alongside FAIR data standards. This ensures "Nothing about us without us" by giving Indigenous groups sovereignty over their linguistic and ecological knowledge. 3. The "AI Labour" Force The millions of workers in the Global South performing data labelling and content moderation—the "human-in-the-loop"—are rarely at the high-level table. The Gap: Discussions focus on "future of work" for white-collar roles while ignoring the current labour rights of the AI supply chain. Inclusion Strategy: Create a dedicated Labour & Human Rights track within the Dialogue that includes international labour unions and digital worker collectives.

1. The "Governance Sandbox" Simulations Instead of debating abstract principles, stakeholders should engage in Policy Stress-Tests. Format: Mixed teams of regulators, AI developers, and civil society members are given a hypothetical "emerging risk" scenario (e.g., a breakthrough in autonomous bioweapon synthesis). Goal: Each team must attempt to apply current international frameworks to resolve the crisis. Outcome: This identifies "legal lag" and interoperability gaps in real-time, providing the Scientific Panel with empirical data on where governance fails. 2. "Reverse Pitching" for Capacity Building To flip the traditional power dynamic, the Dialogue should feature Global South "Problem-Statements." Format: Representatives from developing nations "pitch" specific local challenges (e.g., AI-driven water scarcity or linguistic erasure) to a panel of Big Tech providers and donors. Goal: Instead of tech firms pushing products, they must compete to offer the most open-source, resource-efficient solutions that meet the user's sovereign needs. 3. Distributed "Citizen Assemblies" To address the "UN bubble" problem, the Dialogue should use Hybrid-Synchronous Deliberation. Format: Parallel "satellite" assemblies in regional hubs (e.g., Nairobi, Santiago) are linked via a shared digital platform. Goal: Citizens from the "AI supply chain" (content moderators, data labellers) provide live testimony that is fed directly into the high-level plenary. Outcome: This ensures that the social and ethical implications discussed are grounded in the lived reality of those most impacted by AI deployment.

1. Regulatory "Super-Compliance" Frameworks The Singapore Model AI Governance Framework (2026 Update): This is the first major national framework to specifically address Agentic AI. It offers a "how-to" guide for ensuring human accountability when AI systems take autonomous actions, such as executing financial trades or managing logistics. EU AI Act Compliance Mapping: Platforms like Holistic AI and Lumenova now provide "article-level" mapping. They allow organizations to automate the generation of Conformity Assessments, turning 100+ pages of regulation into a real-time compliance dashboard. 2. Technical "Safety Gate" Platforms AI Safety Institutes (AISI) Network: The collaboration between the UK, US, and Japan AISIs has produced shared Risk Thresholds. These are "red lines" for frontier models (e.g., specific capabilities in biological synthesis) that trigger an immediate governance review before public release. Browser-Native Governance: Tools like LayerX and Harmonic Security address the "Shadow AI" problem by placing governance directly in the browser. They can redact sensitive data (PII) in real-time before a user pastes it into a public LLM, ensuring data protection without blocking innovation. 3. Open-Source Transparency Tools Egeria (Linux Foundation): An open-source metadata standard that allows different governance tools to "talk" to each other. It ensures interoperability by creating a common language for tracking data lineage and model versions across different vendors. VerifyWise: A new open-source platform that provides a "Model Registry." It allows smaller organizations to catalogue their AI assets and perform risk scoring using the same benchmarks as global enterprises, preventing a "governance gap" between Big Tech and SMEs.