Earlier this month, Geneva hosted the inaugural UN Global Dialogue on AI Governance, the AI for Good Global Summit, and the WSIS Forum 2026, convening 163 countries along with business and civil society. We took the pulse in the rooms. This issue of The Brief highlights the signals, intelligence, and new research from Geneva. 

Overall, the convenings produced broad high-level agreement on governance gaps and priorities, while also showcasing case studies, research, and innovations.  Underneath, consensus on a fundamental question went unresolved: governance toward what end?  The clearest answer came from Mary Robinson, former President of Ireland. Speaking to the balance of governance, innovation, and safety, she reminded delegates twice that “a government’s first responsibility is to protect its citizens.”

From heads of state to civil society leaders across the globe, stakeholders emphasized imperative to operationalize governance principles quickly, and researchers brought evidence on what works. Yet for the most part, the discussion stayed diplomatic and high level, with limited attention to building the connective tissue that turns principle and evidence into working governance systems. Building this “missing middle” is an urgent priority for leaders.

 

On the Radar: Strategic Signals

 

July 6–10, 2026  |  Global Meetings, Geneva

The inaugural UN Global Dialogue on AI Governance convened representatives from over 170 member states, along with business, civil society, and academia, with the goal of ensuring that AI governance reflects the priorities of all nations and that the benefits of AI are shared by all. It was designed as an inclusive platform for discussion among governments and other stakeholders, organized around four themes:

• AI opportunities and implications
• Bridging AI divides
• Safe, secure, and trustworthy AI, and interoperability and compatibility of approaches
• Respecting, protecting, and promoting human rights

The Dialogue surfaced urgent priorities around:

• Addressing the gap between technological advancement and governance
• Tailoring governance to the moment by adopting “dynamic governance” approaches able to respond to a rapidly evolving context
• Addressing critical weaknesses in institutional and human capacity to govern and absorb AI
• Developing interoperable governance frameworks among nations and jurisdictions
• Closing gaps in the research and evidence needed to guide governance and to understand what works and its impact
• Integrating rights- and justice-centered approaches and safeguards into governance systems
• Adopting dynamic, systemic, and life-cycle and impact focused safety testing and assessment
• Expediting the “build-out” of governance and evaluation bodies

The AI for Good Global Summit, convened by the ITU, ran alongside the Dialogue with the goals of identifying practical, scalable applications of AI that advance the Sustainable Development Goals and to connect AI innovators with UN agencies and governments. It launched the AI for Good Global Commission, convening more than 40 government, business, and UN leaders to identify practical pathways to advance trust, access, and responsible innovation. As constituted, the body omits civil society, a gap that has already raised concerns and calls for collaboration. The summit also established a multilateral Agentic AI Focus Group with a mandate on trusted digital identity and life-cycle accountability for AI agents.

Our takeaways

• The governance-capability gap is immense and global. Stakeholders across all three venues warned that AI is advancing faster than governance capability. The Dialogue surfaced both institutional and human capacity gaps in governing and absorbing AI across developing and advanced economies alike, though the gap is not uniform and varies substantially among geographic regions. Strengthening this capacity was flagged as an urgent priority. The near-term governance environment will remain highly fragmented and volatile as nations work to address these gaps. 
• Dynamic governance is an emerging model. Governments converged on the need for adaptive governance capable of responding to the evolving context. There was also consensus around the imperative for continuous, systemic, and life-cycle monitoring and impact assessment. These priorities are likely to translate into dynamic compliance regimes.
• Interoperability among governance frameworks is a priority to watch. Multiple high-level contributors urged action to advance interoperability of governance frameworks across borders. Though various models for operationalizing this were discussed, no specific principles or mechanisms were adopted.
• Human rights and accountability are priority areas for action, with more frameworks emerging to support them. Opening the Dialogue, the UN Secretary-General emphasized the need to prioritize human rights dimensions of AI transformation, noting that “AI must never strip away dignity or entrench discrimination,” and that “in every high-stakes decision, in justice, in healthcare, in policing, machines can inform, but humans must decide, and answer.” Civil society groups released new frameworks for supporting this on the sidelines of the event.
• Systems-level agent accountability. The summit highlighted research and empirical case studies on agentic governance, revealing shortcomings in existing frameworks for governance, legal regulation, and testing and monitoring approaches. It also established new mechanisms for deeper collaboration in understanding and governing agentic risks.
• Consensus that full control is not attainable. A recurring theme was that even with full deployment of governance and safety measures, complete risk mitigation and control of agentic systems is not possible. Many urged new multilateral frameworks for standards and governance, pointing to potential models such as nuclear non-proliferation regimes and World Meteorological Organization, as well as mechanisms to support coherent responses to emerging developments and crises, such as “red phones for the AI age.”
• Discussions concentrated on topics where there was political space around geopolitical sensitivities, leaving gaps. Across heads of state, corporations, and civil society there was much cohesion and consensus, at least in the public sessions, on priorities. That cohesion, however, was largely high level and did not extend to specifics or operational details. As the conveners and others work to negotiate and shape operational steps in the months ahead, numerous gaps and differences remain to be bridged.  Leading inclusive processes to resolve these where civil society and diverse geographies are full participants is imperative but is likely to be a challenge.  On what was not discussed, see Konstantinos Komaitis, The Internet Had a North Star. The UN’s Global Dialogue Made Clear AI Doesn’t., Tech Pol’y Press (July 15, 2026).

→  U.N., Global Dialogue on AI Governance (last visited July 20, 2026).

→  Inaugural UN Global Dialogue on AI Governance Ends with Call to Turn Principles into Action Before 2027, Digital Watch Observatory (July 8, 2026).

→  Press Release, Int’l Telecomm. Union, Global Leaders Launch AI for Good Global Commission (July 2, 2026).

→  Konstantinos Komaitis, The Internet Had a North Star. The UN’s Global Dialogue Made Clear AI Doesn’t., Tech Pol’y Press (July 15, 2026).

The WSIS Forum

The WSIS Forum 2026 convened the World Summit on the Information Society, the UN’s foundational process on digital development, with a focus on operationalizing the WSIS action lines (ICT infrastructure, capacity building, cybersecurity, and enabling environments) and aligning them with the Sustainable Development Goals. Sessions addressed closing digital divides, countering misinformation, advancing equality, and sustaining a multistakeholder model for internet governance and digital development. Several sessions took up data justice and the alignment of AI governance with human rights.

Our takeaway. WSIS, and the Forum, anchors AI governance in an established multilateral architecture built around digital development and human rights. Its objectives of advancing people-centered, inclusive, development-oriented information society: bridging the digital divide so everyone can access, use, and share information and action-line implementation orientation is precisely the practical missing middle the Dialogue lacked; tighter integration between the two could convert principles into operational governance.

→  Jhalak M. Kakkar, Jason Pielemeier & Elonnai Hickok, As the UN Launches Its Global Dialogue on AI Governance, WSIS Offers Critical Lessons, Tech Pol’y Press (July 2, 2026).

→  IT for Change, Putting Data Justice at the Heart of AI Governance, WSIS Forum 2026, Session 385 (July 8, 2026).

→  Data Privacy Brasil, How Can We Articulate Global AI Governance with Human Rights and Multistakeholder Engagement?, WSIS Forum 2026, Session 424 (July 9, 2026).

The Independent Evidence Base

The week included a wave of independent research releases. Alongside the scientific assessment and measurement platforms, some of the most concrete work came from civil society and independent researchers advancing rights and justice.

Independent International Scientific Panel on AI, Preliminary Report

The Independent International Scientific Panel on AI, seated by the UN General Assembly in early 2026, released its Preliminary Report, the first global scientific assessment of AI’s opportunities, risks, and impacts. It finds that leading systems now double their capability to complete software tasks every four to seven months and states the core governance dilemma plainly: policymakers need evidence to act, yet, as the Panel puts it, “by the time the evidence exists, it might be too late.” It warns that the widening gap between rapidly improving capabilities and effective risk-management methods could produce catastrophic outcomes. On AI agents, the Panel cautions that there are no known technical guarantees that agent systems will follow their instructions, and that evidence of systems acting contrary to instruction is already accumulating.

Our takeaway. The Preliminary Report’s core message is a structural timing problem: capability compounds on a scale of months while the methods to manage risk lag. Critically, on agents, the Panel is explicit that no technical guarantee yet ensures that autonomous systems will follow instructions. The report is both warning and tool, a reference leaders can use to anchor governance arguments in empirical science.

→  Indep. Int’l Sci. Panel on AI, Preliminary Report (July 1, 2026).

→  ‘The Science Is Here’: UN Chief Welcomes First Global AI Assessment, UN News (July 1, 2026).

Partnership on AI, Global Responsible AI: Measures of Progress

The Partnership on AI announced two instruments. One is a Global AI Progress Hub, a cross-sector platform that lets organizations document responsible-AI actions and track impact on wellbeing, employment, and the economy. A second is an independent annual report, Global Responsible AI: Measures of Progress, guided by a multistakeholder community. Both instruments target post-deployment measurement assessing how systems perform in operation, not what they promise before release.

Our takeaway. The initiatives promise to facilitate the evolution of pre-deployment assessment to life-cycle monitoring and evaluation. The Partnership on AI’s instruments are built to continuously track what systems actually do in operation, across wellbeing, employment, and the economy, and shift the evaluation landscape from a narrower compliance-centered focus to continuous auditing and impact assessment.

→  Press Release, P’ship on AI, Partnership on AI Announces New Global Initiatives to Measure Progress in Responsible AI (July 6, 2026).

Global Index on Responsible AI 2026

The Global Center on AI Governance launched the Global Index on Responsible AI 2026 in Geneva on July 8, assessing 135 countries and jurisdictions against more than 68,000 data points across five dimensions: Inclusion and Diversity, Ethics and Sustainability, Labor and Skills, Trust and Safety, and Use of AI in Public Service Delivery. Its central finding is an implementation gap: evidence shows that among countries where AI governance frameworks exist active implementation is underway in only 55% of cases. In the Global Majority countries active implementation is even lower at 45%.  It’s key findings:

• AI is accelerating faster than governments can govern it in the public interest
• Responsible AI governance is expanding in the Global South, but binding protections remain scarce
• AI safety is governed as a technical problem, while human harms stay under-addressed
• Governments regulate AI transparency but do not disclose their own use of AI
• Gender is increasingly recognised in AI governance, but protection from gendered harms remains weak
• Future generations are prepared for the AI economy but not protected from AI-related harms
• AI’s environmental footprint remains a blind spot in responsible AI governance
• Governments recognise the need for local-language AI but do not require developers to deliver it
• Governments are investing in AI skills but neglecting workers’ rights
• Global AI governance is fragmenting before a shared floor of protection is established

 The report found that governments move faster on capability than on protection: 53% have AI skills-training measures but only 29% protect workers, just 27% address AI’s environmental impact, and only 18% require disclosure of the AI systems governments themselves deploy. Use of AI in public service delivery, the dimension touching welfare, healthcare, education, and policing, scores lowest of the five. It makes several recommendations for addressing these gaps and broader findings. 

Our takeaway. The Index illuminates the magnitude of the implementation gap highlighted at the Dialogue with data.  It also provides detailed context, insights, and guiding principles for closing these gaps and broader operational work of the Dialogue process and beyond.  

→  Glob. Ctr. on AI Governance, Global Index on Responsible AI (2d ed. 2026).

Feminist Guiding Principles for Global AI Governance

On the Dialogue’s opening day, the Gender in Digital Coalition, whose members include UN Women, Equality Now, and the Association for Progressive Communications, released the Feminist Guiding Principles and Commitments for Global AI Governance. The principles spotlight a status quo where AI systems are not neutral, because they are designed, trained, deployed, evaluated, and governed within institutions that already reflect unequal power. As a result, women, girls, and gender-diverse people face disproportionate AI-related harms while remaining underrepresented in the bodies that build and regulate these systems. The document makes the case for embedding feminist commitments from the outset, with “representativeness” serving as a governance requirement across the AI lifecycle and as a continuous measure of real-world impact.

Our takeaway. The principles offer concrete practical strategies for advancing inclusion and equality from design through deployment and beyond.  Its framework shifts equity from an abstract compliance goal to a core metric that can be integrated in evaluation, procurement, and impact assessment.

→  Gender in Digit. Coal., Feminist Guiding Principles and Commitments for Global AI Governance (July 6, 2026).

Africa AI Governance Index (AAGI) 2026

Lawyers Hub and The Africa AI Policy Lab released a pioneering African continent-wide mapping of AI governance readiness. The index is Africa designed and led and analyzes 54 countries. Key findings include that:

• Readiness across the continent is early-stage
• Data-center capacity concentrated
• AI talent is scarce
• Dependence on foreign infrastructure is high
• Enforcement of AI laws is limited
• The pace of Investment is exceeding safety
• AI workforce protections are lacking
• African-language AI content is almost zero

Our takeaway. This landmark index closes a pivotal gap in Africa-led research and assessment on AI governance. Grounded in continental frameworks, it evaluates the full governance chain across strategy, regulation, infrastructure, skills, ethical safeguards, and regional cooperation and provides vital evidence for governments, researchers, and civil society in leading transformation that benefits its people.

→  Africa AI Governance Index (2026)

HumRights-Bench

HumRights-Bench, described as the first expert-validated benchmark for whether AI systems can reason about human rights grounded in international human rights law, was presented to UN Member States at the Human Rights Council shortly before the global meetings. It was developed by AI & Equality at Women at the Table with researchers from Hunter College, the Oxford Internet Institute, King’s College London, Georgetown University, and the University of Oslo. The bench is designed to assess model capability to identify rights violations, the applicable law, analyze and propose a remedy.  When frontier models, including GPT-5, Claude Opus 4.7, and Gemini-3, were tested on their ability to perform these tasks on fact pattern involving the right to water scores ranged from 51% to 58%.  Models received the lowest scores when tested for their ability to recognize when a right existed. Bench leaders have indicated that an expanded version covering additional rights such as due process and education will be released in coming months. 

Our takeaway. HumRights-Bench supplies novel testing and performance data on model reasoning about rights. The results of its initial tests highlight the urgent need for further testing and benchmarking in this area and for building the evidence base on protecting and advancing human rights.

→  AI & Equality (Women at the Table), HumRights-Bench (last visited July 20, 2026).

Content Provenance and the Surveillance Risk

WITNESS launched C2PA Content Credentials and the Surveillance Risk, examining the content-authenticity and provenance systems now promoted to counter AI-generated misinformation. Through seven adversarial scenarios, it found that provenance tools, absent safeguards, can enable identity disclosure, behavioral profiling, and surveillance, and that technical fixes alone cannot secure a trustworthy information ecosystem. Its conclusion is that governance must evolve alongside the technology to protect privacy, civic space, and the people who create and document content.

Our takeaway. Even the tools built to restore trust can bring risks. WITNESS shows that content-provenance systems, absent deliberate safeguards, can enable identity disclosure, profiling, and surveillance.  Governance and rights safeguards be designed into system infrastructure. Continuous testing and impact assessment is crucial as systems evolve. 

→  WITNESS, C2PA Content Credentials and the Surveillance Risk (July 8, 2026).