Mennonite Central Committee

A successful first Global Dialogue on AI Governance should land a few practical outcomes. First, it needs to clearly surface what AI actually looks like on the ground outside high-income, high-infrastructure environments, especially in humanitarian, development, and nonprofit contexts. That means being honest about power imbalances between funders, implementers, and the communities affected, along with real constraints like limited connectivity, lower digital literacy, and weak data protection. Second, it should move past high-level principles and get into shared problem definitions and next steps. Rather than trying to solve everything or reach broad consensus on regulation, the focus should be on identifying a small set of priority gaps, like accountability for AI-related harm, data protection in low-resource settings, and how non-state actors are using these tools, and then outlining how to test, learn, and build guidance together. Third, the process itself has to feel legitimate. That means real participation from civil society, humanitarian groups, and affected communities, not just being consulted after the fact. There should be clear commitments to ongoing engagement, transparency around disagreements, and a path to carry this work into future intergovernmental discussions so it does not just end here. Long story short, success is not about producing a polished document. It is about whether people walk away with a clearer sense of shared responsibility, realistic expectations, and actual opportunities to work together across sectors and regions

• Safe, secure and trustworthy AI
• AI capacity-building
• Protection and promotion of human rights
• Transparency, accountability, and human oversight

These priorities come out of what we are actually seeing in real-world AI use across very different and uneven environments. Safe, secure, and trustworthy AI is the baseline, especially in places where regulation is limited and institutional capacity is low. In those contexts, the risks of misuse, data exposure, or unintended harm are much higher. Without those basic safeguards in place, it is hard to justify or sustain any of the potential benefits. Capacity-building is just as urgent. A lot of organizations and communities are already using AI tools, often informally and without much support. This is not just about technical training. It also needs to include governance literacy, so people understand the risks, accountability, and how to make informed decisions in their specific context. Protecting and promoting human rights is non-negotiable, particularly where AI intersects with vulnerable populations. Issues like consent, surveillance, bias, and exclusion tend to be more pronounced in humanitarian and development settings, where people may not have a real ability to opt out or challenge decisions. Transparency, accountability, and human oversight are also key to maintaining trust. In practical terms, that means being clear about when and how AI is used, ensuring humans remain responsible for decisions influenced by AI, and having ways to respond when something goes wrong. Taken together, these priorities lean toward responsible use over rapid adoption. They also highlight governance gaps that are already showing up in real situations today.

One cross-cutting issue that needs more attention is how AI is governed in relationships with clear power imbalances, especially between funders and grantees, employers and workers, and organizations and communities. A lot of existing frameworks assume informed consent and equal footing, which does not reflect how these relationships actually work in practice. Another emerging challenge is data governance in low-infrastructure settings. Questions around data ownership, stewardship, localization, and long-term risk get complicated quickly where legal protections are weak or not enforced. Many AI governance conversations assume strong national data protection frameworks that simply are not there in a lot of contexts. Environmental sustainability is also underdeveloped. There is growing awareness of AI's environmental impact at a high level, but very little practical guidance for organizations that are already resource-constrained and trying to balance environmental costs with potential social benefits. Finally, there is a widening gap between global AI governance frameworks and what organizations can actually use day to day. There is a missing middle where high-level principles do not translate into practical guidance. Closing that gap means focusing more on operational realities, not just policy design.

In the nonprofit and humanitarian sector, governance gaps are already shaping how AI is being adopted, often in fragmented and inconsistent ways. Organizations are being pushed to use AI to improve efficiency and reporting, but do not have clear guidance on acceptable use, accountability, or data protection, especially when working with partners in low-resource settings. Data protection is a major challenge. When governance is weak or uneven, the risk of sensitive data being mishandled goes up, particularly if staff are using freely available AI tools without strong safeguards. On the flip side, overly restrictive or unclear policies can make people hesitant to experiment or learn where there could be real value. There are also real opportunities here. More practical and consistent governance could support safer innovation, make it easier to share learning across organizations, and improve access to AI benefits more equitably. Capacity-building and shared standards can reduce duplication, lower risk, and help align AI use with humanitarian principles and human rights. So this is not just about managing risk. It is also about creating the conditions for responsible use that actually supports mission-driven work.

The AI Dialogue can act as a bridge between global policy conversations and what is actually happening on the ground. Its value is in creating space for cross-sector learning, surfacing governance challenges that cut across regions, and building alignment without forcing one approach on everyone. By bringing together governments, civil society, technical experts, and affected communities, it can help build a shared understanding of risks and responsibilities, especially for non-state actors who are often missing from formal regulatory processes. It can also support more practical cooperation through shared principles, real-world examples, and pilot efforts that feed into future governance work. Most importantly, it can help elevate voices that are currently underrepresented, so international AI governance better reflects a wider range of experiences and needs.

The AI Dialogue should connect with existing efforts like the Humanitarian AI Code of Conduct, NetHope's AI governance research, and other emerging organizational-level frameworks that are trying to turn high-level principles into something practical. These initiatives bring real insight into issues like consent, accountability, and data protection in complex environments. Where the Dialogue adds value is in linking this work to intergovernmental processes. That includes amplifying what is already being learned, identifying shared gaps, and avoiding unnecessary duplication. It can also act as a neutral space to coordinate across efforts that are currently running in parallel.

Different stakeholders can contribute through case studies, lived‑experience testimony, technical analysis, and policy reflection. To support this, the Dialogue should combine plenary sessions with smaller, facilitated discussions focused on concrete governance challenges. Participation should be supported through multilingual materials, remote access options, and clear expectations about how input will be used. Transparency about outcomes and follow‑up is essential to maintaining trust and engagement.

Underrepresented voices include organizations and communities operating in low‑income or crisis‑affected settings, local implementing partners, and practitioners responsible for day‑to‑day AI use rather than policy design. Inclusion requires more than invitations. It might require resourcing participation, valuing experiential knowledge, and creating formats that do not privilege technical or legal expertise alone. This may include regional consultations, stipends, and partnerships with trusted networks like Nethope.

Scenario-based discussions, governance "clinics," and facilitated peer exchanges are likely to drive more meaningful engagement than standard panels. These formats give participants a chance to work through real dilemmas, think through trade-offs, and learn from different perspectives in a practical way. It is also worth thinking about how the Dialogue is documented. Capturing not just areas of agreement, but also where there is disagreement, would make the outputs more useful and grounded.

Promising practices include organization-level AI policies that are grounded in values, along with clear approval and disclosure processes, and a strong focus on human accountability and oversight. Capacity-building efforts that bring together ethics, governance, and technical training tend to be especially effective. There is also value in shared, collaborative platforms that offer templates, risk assessments, and lessons learned. These can lower the barrier for smaller organizations that do not have the same resources. These approaches show that effective AI governance is achievable when the guidance is practical, grounded in context, and aligned with the organization's mission.