I. Assessing the UN’s First Steps on AI Governance
The inaugural UN Global Dialogue on AI Governance (UNGDAI) took place on July 6–7, 2026, in Geneva. Mandated by the Global Digital Compact, the UNGDAI brought together over 2,000 regional and sectoral stakeholders to address key AI issues through an agenda built on four core pillars: harnessing AI opportunities; bridging digital divides; promoting safe and trustworthy AI; and, critically, advancing human rights.
The Co-Chairs of the UNGDAI – the Permanent Representatives of El Salvador and Estonia to the UN – with support from a Secretariat drawn from relevant UN agencies and offices, deserve credit for putting together a process that delivered consultation, thematic coherence, and a well-organized event in just over six months. Two major successes of the UNGDAI were its commitment to inclusivity and its centering of human rights, both of which were priorities championed by civil society during consultations. Despite this good work, a glaring lack of clarity remains regarding the purpose of the UNGDAI and its relationship to other AI governance processes, such as the recent India AI Impact Summit and next year’s Swiss AI Summit.
The Co-Chairs’ will write a “summary” from this first dialogue, which should help frame the process for continued consultations and the focus for the agenda of the second UNGDAI, which will be held in New York in May 2027. Together with the reports of the Independent International Scientific Panel on AI (Scientific Panel), the UNGDAI discussions and summaries will set the terms and context for “intergovernmental consultations” ahead of the General Assembly’s “high-level review” of the UN Global Digital Compact at the end of next year.
II. Grounding AI Governance through Evidence & Diverse Perspectives
Most AI governance processes, including the UNGDAI, remain focused on awareness-raising and hand-wringing. While not without its limitations, the Scientific Panel’s preliminary report helpfully highlights several key trends, gaps, challenges, and opportunities and explains the current state of the technology. Structured across seven key domains, ranging from technical trajectories and societal impacts to human rights and governance, the panel’s findings can be used to identify key focus areas that can be further explored through multistakeholder input and advanced through AI governance processes. Crucially, these evidence-based insights should be further contextualized and ground-truthed at national and regional levels in order to provide a feedback loop between global processes and a diversity of lived realities, as well as to facilitate coordination and interoperability across policy efforts.
The Global Network Initiative and the Centre for Communications Governance, NLU Delhi launched the MAP-AI project in October 2025 to enable multistakeholder approaches and participation in AI governance, with an emphasis on the Global Majority. Through MAP-AI, we have been identifying and articulating bottom-up, global majority-driven priority areas for AI governance by facilitating convenings, insights, and collective input.
Based on discussions across the MAP-AI community and drawing on some of the key insights from the Scientific Panel’s preliminary report, we have identified an initial, non-exhaustive list of key policy areas that can be advanced through international AI governance processes in order to drive concrete, actionable outcomes. We have structured these policy focus areas around the three central pillars that have guided MAP-AI’s work: Safe & Trusted AI (encompassing assessments, cybersecurity, and incident reporting), Global Governance (focusing on institutions, corporate accountability, and regulatory frameworks), and Context-Driven Infrastructure (prioritizing multilingual AI and regional capacity, capability, and access). Cutting across these pillars is the underlying international human rights framework, which serves as a universal normative baseline that helps ensure processes are built on due process and participatory principles, provides normative coherence to help identify priorities, and guides policy interventions to ensure transparency and accountability and an emphasis on Global South leadership. These policy areas can be mapped onto the core themes and potential working group structures of various international AI governance processes, including the UNGDAI and the Swiss AI Summit.
Over the coming months, we hope to refine these focus areas, identify existing work to build on, and develop new actionable policy proposals that can be advanced through international AI governance processes.
A. Safe & Trusted AI
Assessments, evaluations, benchmarks, standards & thresholds: The Scientific Report notes that evaluation methods, benchmarks, assessments, and thresholds for AI systems are underdeveloped. Such mechanisms need to be dynamic, enable systematic evaluation and monitoring, and be defined and assessed by independent stakeholders outside of the private sector.
Policy proposals can further explore contextualizing risk by defining “unacceptable risk” through regional human rights, legal, and socio-political lenses; establishing robust enforcement mechanisms to hold actors accountable to standards; incentivizing independent stakeholders outside the private sector to assess systems; and defining resourcing and capacity-sharing mechanisms to help regions overcome domestic constraints.
Cyber Security: The Scientific Panel’s report notes that AI enables more sophisticated cyberattacks and changes the economics of cybersecurity. It reduces the costs of offensive capability and creates systemic risks, while agentic systems complicate attribution and accountability. Meanwhile, increasing reliance on private-sector infrastructure blurs traditional distinctions and the balance of power between public and private actors.
Policy proposals can further explore ways to use AI to upgrade legacy infrastructure, collaborative mechanisms for rapid responses to critical incidents, establishing safe harbors to protect safety research, and standardized transparency and incident disclosure expectations. Furthermore, policy proposals can establish mechanisms for cross-border cooperation, including sharing threat intelligence and attribution.
Incident Reporting: The Scientific Panel’s report notes the need for continuous measurement, including tracking how a system behaves after release in real-world settings, with real users, real tasks. Such monitoring can take place through incident reporting databases and can include anonymized, aggregated AI usage patterns provided by AI developers, as well as user-reported outcomes.
Policy proposals can further explore key design, governance, and contextualization considerations for incident reporting. This includes adopting standardized data taxonomies, such as those set out by the OECD and the AI Incident Database, while establishing independent governance structures. Localized risk classifications (e.g., electoral manipulation, land rights), severity ratings that account for relevant capacity constraints, and formal structures for integrating community-led evidence and qualitative human impact stories need to be explored.
B. Global Governance
Corporate Accountability: The Scientific Panel’s report highlights that technological development is outpacing the evidence needed to make informed and effective policy decisions with respect to AI. Frontier AI capabilities are concentrated in a handful of companies and countries, while critical decisions on training data, safety, deployment, and model access remain with the private sector and largely outside public oversight. The result is an information gap between governments, industry, civil society, and the public.
Policy proposals to enhance corporate accountability should build on the UN Guiding Principles on Business and Human Rights (UNGPs) and can include standardizing company-to-government disclosure templates for security incidents, malfunctions, and misuse, as well as establishing clear categories for public transparency reporting. Additionally, efforts should be made to establish jurisdiction, region, and/or language-based risk factors and processes for consolidating and making available credible data, including from civil society and public reporting, for use in ongoing corporate due diligence and risk management. Institutionalizing inclusive stakeholder engagement mechanisms, such as external advisory boards and grievance channels that involve local civil society, labor representatives, and affected host communities, can help ensure responsible AI development and deployment.
AI Governance Institutions: The Scientific Panel’s report notes that establishing national and regional AI safety institutes can build capacity at the national and regional levels and provide independent capability and risk assessments. Yet, these institutions remain embryonic outside a few, powerful countries.
Policy proposals should help clarify the range of mandates, resource requirements, and governance of AI Safety Institutes, prioritizing the preservation of institutional independence, the public interest, and integration of civil society input. Practical and collaborative models for supporting safety research and effective oversight of AI deployment in under-resourced contexts, will also be important to explore.
Regulatory frameworks and government powers: The Scientific Panel’s report highlights that dozens of distinct governance instruments exist across countries and contexts. However, these are concentrated in a few key jurisdictions, rely heavily on cooperation from the companies whose conduct they seek to control, and rarely measure and report transparency on their effectiveness. There is a need to explore mechanisms to enable access to these companies for jurisdictions across the world.
Besides this, policy proposals should examine how existing privacy, data protection, and consumer protection laws are being used to govern AI. Robust data governance frameworks and collective data rights should be explored as part of safeguarding local culture and data. Policy proposals can also explore which state powers for emergency interventions (e.g., mandatory safety pauses or rollbacks) could be considered appropriate and how their use can be subject to checks and balances to prevent government overreach.
C. Context-driven Infrastructure
Multilingual AI: The Scientific Panel’s report notes that most of the world’s languages and cultures remain underserved by AI. This not only creates a risk of exclusion and homogenization, but also a safety risk. Diverse AI systems, support for public datasets, and benchmarking initiatives for underrepresented linguistic and cultural contexts are needed.
Policy proposals can establish sustainable funding mechanisms, such as compute grants and regional research consortia, to maintain local datasets alongside community-managed data trusts that keep public data accessible. Additionally, adapting copyright frameworks to protect local knowledge ecosystems and developing public-interest benchmarking initiatives to evaluate AI models for linguistic fluency, cultural nuance, and contextual safety need to be explored.
Capacity, capability, and access: The Scientific Panel’s report notes that access to AI tools alone does not yield equal benefit; complementary investments in data, skills, workflows, and institutions that enable effective, cost-effective, and safe deployment are equally necessary but unequally distributed.
Policy proposals can further explore building shared regional compute hubs that use independently certified hardware and software to facilitate local research, innovation, and public-sector use cases, supported by renewable energy and energy-efficiency practices. Concurrently, incentives can be explored to promote open-weight, small-parameter models optimized for low-bandwidth networks, paired with public sector training to equip local policymakers, judges, and civil servants with the skills needed to effectively evaluate, procure, and use AI.
III. The Path Forward
The UNGDAI proceedings in Geneva represented an important step towards inclusive AI governance. Critical focus now shifts to the intersessional work, the 2027 UNGDAI convening in New York, and the forthcoming Swiss AI Summit. The policy ideas set out above respond to the evidence and framework for action set out by the Scientific Panel, stem from conversations across diverse stakeholders and regions, including the Global Majority. While establishing global AI governance remains a complex, multi-stage endeavor, these ideas can help move the conversation forward from principles to practical and concrete initiatives that respond to real needs and impacts. This blog is meant to get that conversation started. Over the coming months, we will continue to refine and develop concrete policy proposals that can be advanced through AI governance.