FAO

A successful Dialogue would deliver practical, inclusive and implementable outcomes that improve people's lives while protecting rights and ecosystems, including for rural communities and Indigenous Peoples. For agrifood systems, success would include: (1) a shared understanding of responsible AI principles tailored to food and agriculture—safety, transparency, human oversight, sustainability, and accountability; (2) clear capacity-building pathways so low- and middle-income countries, farmers' organizations, and extension services can adopt AI responsibly and benefit equitably; (3) commitments to interoperable governance approaches to reduce fragmentation and enable cross-border collaboration on data, early warning, and research - recognizing that responsible AI rests on high-quality, interoperable, well-governed data (including official statistics) and that strengthening national data and statistical systems in low- and middle-income countries is a precondition, not a parallel track, for inclusive AI governance- . and (4) concrete actions to expand responsible access to open-source tools, open data and open AI models where appropriate, while safeguarding privacy, biosecurity and commercial/trade sensitivities. The Dialogue would be successful if it catalyzed partnerships (governments, UN entities, private sector, academia, civil society) around 5 missions (i) closing the field intelligence gap; (ii) unlocking fair and efficient market access; (iii) building risk-ready agrifood systems; (iv) upgrading institutional capacity; and (v) optimizing farming through analytics, as well as key use cases—e.g. early warning for food insecurity, climate-resilient advisory services, animal/plant disease surveillance, and supply-chain transparency—paired with proportionate risk and impact assessment for high-impact agrifood AI (purpose, impacts, key risks, mitigations, monitoring) and clear accountability, as outlined in FAO's Digital Agriculture and AI Innovation Roadmap (https://doi.org/10.4060/cd5956en), including the AI governance toolkit. Finally, success would be demonstrated by agreed next steps: a timebound workplan, mechanisms to include voices from the Global South and rural communities, and a simple approach to track progress (e.g., inclusion, safety incidents, capacity development) between sessions.

• Safe, secure and trustworthy AI
• Interoperability of governance approaches
• Transparency, accountability, and human oversight
• Open-source software, open data and open AI models

These priorities align with using AI to accelerate agrifood systems transformation while managing risks that can disproportionately affect smallholders and vulnerable communities, consistent with a human-centred responsible AI approach. • Safe, secure and trustworthy AI: Agrifood applications increasingly inform decisions on inputs, finance, markets, insurance, and emergency response; failures or misuse can cause real-world harm. Assurance approaches-risk/impact assessments, local validation, and safeguards against manipulation-are essential. -Transparency, accountability and human oversight: Agrifood AI applications often influence decisions affecting livelihoods, access to services and resource allocation. Ensuring explainability, clear accountability arrangements and meaningful human oversight is essential, particularly in public-sector and humanitarian contexts. FAO contributes by translating these principles into operational guidance, institutional processes and monitoring mechanisms that can be applied by ministries of agriculture, extension services and partner institutions. • Interoperability of governance approaches: Food and agriculture issues are cross-border (trade, pests/diseases, climate risks). Interoperable standards and compatible approaches can enable responsible data sharing, joint early-warning systems, and scalable solutions, aligned with digital public goods where appropriate. • Open-source software, open data and open AI models: Appropriately governed openness can underpin a more robust information ecosystem, reduce costs, increase transparency, and enable adaptation to local languages and agroecological conditions. Openness should be paired with data governance (privacy, consent, Indigenous/community data considerations, statistical disclosure control and the protection of microdata), quality assurance, and accountability for downstream use, including metadata/provenance and clear intended use/limitations. Together, these priorities position FAO as a bridge between global AI governance principles and practical, sector-specific implementation in agrifood systems, grounded in country experience and focused on delivering safe, inclusive and accountable outcomes.

From a food and agriculture perspective, several cross-cutting issues merit attention: 1) Data governance for agrifood systems: Beyond openness, governance should address consent, privacy, data quality, and rights over rural communities and Indigenous data, including benefit-sharing and avoiding extractive data practices. Governance should be grounded on the principles of value (maximizing the value of data across all data domains through increased responsible data use and reuse), trust (ensuring secure environments for data across the data life cycle and by protecting individuals and groups from the risks and harms arising from data misuse) and equity (ensuring equitable distribution of benefits from increased and responsible access, use and reuse of data). In addition, an openness framework that includes open-source software, open data and open AI models is incomplete if it does not also take into consideration Open Access of written content, which is often a major component of training corpora. 2) Environmental and climate impacts of AI: Compute, energy and water footprints-and the lifecycle impacts of digital infrastructure-should be considered alongside AI's potential to support mitigation and adaptation in agriculture. 3) Inclusion and last-mile delivery: Treat inclusion as a design requirement-participatory design with affected communities, low-connectivity modes, and adaptation to local languages and literacy levels-while recognizing extension services as trusted intermediaries. 4) Market power and dependency risks: Concentration of data, models and platforms can create lock-in and reduce national/community agency. Governance should promote portability and fair competition, with options for local hosting where needed. 5) Integrity of information ecosystems: Misinformation and AI-generated content can affect food safety, input markets and crisis response. Governance of knowledge production and dissemination that incorporates guidance around the use of AI, as well as provenance and clear disclosure when AI-generated content is used, are increasingly important. 6) Grievance and remedy mechanisms: People affected by AI-driven decisions (e.g., credit scoring, insurance, targeting of aid) need accessible avenues for explanation, contestation and redress.

In agrifood systems, governance gaps create high-impact risks and missed opportunities. Governance gaps also slow the modernization of national agricultural statistical systems, where AI offers concrete opportunities (e.g., improved survey design, satellite and administrative data integration, cost-effective small-area estimation, and more frequent updates of food security indicators) that depend on clear rules for data sharing across statistical, administrative and private-sector sources. Where transparency, accountability and data governance are weak, AI-enabled advisory tools and automated decision systems can produce harmful recommendations (e.g., inappropriate input use), amplify bias in access to credit/insurance or targeting of assistance, and reduce trust in public services. Weak monitoring and feedback can allow model drift and harm to persist. Fragmented approaches also make it harder to share data across borders for early warning on food insecurity, pests/diseases and climate hazards, and can slow responsible innovation by creating legal uncertainty for public institutions and local enterprises/small businesses. Labour impacts are also a critical dimension of AI governance in agrifood systems: while AI can raise productivity and support new roles in data stewardship, advisory services and system oversight, weak governance risks uneven displacement of routine tasks, opaque or biased workplace decisions, and widening inequalities—particularly for rural workers, women and youth—underscoring the need for human oversight, transparency and skills investment to support a just transition and shared benefits. At the same time, advances in the thematic areas selected—trustworthy AI, capacity-building, interoperability and appropriately governed openness—can unlock major benefits. Stronger safety and transparency practices enable AI to be used more confidently in risk-sensitive settings (e.g., humanitarian response, food safety monitoring). Capacity-building can help countries develop the skills, institutions and infrastructure needed to deploy AI in extension services and national early warning systems while protecting human rights. Interoperable governance approaches can accelerate collaboration on shared challenges such as transboundary pests and market volatility. Finally, open-source tools, open data and open models—paired with privacy, consent, quality standards and biosecurity safeguards—can lower costs, support local-language solutions, and reduce dependency on a small number of providers, helping ensure that benefits reach smallholders and vulnerable communities.

The AI Dialogue can play a pivotal role in advancing international cooperation on AI governance by serving as a trusted multilateral bridge between global principles and sector‑specific, country‑level implementation. For FAO and agrifood systems transformation, this role is particularly critical, as AI increasingly shapes food security, climate resilience, natural resource management, and rural livelihoods across borders. First, the Dialogue can foster convergence and interoperability of AI governance approaches, helping reduce fragmentation across initiatives and enabling countries—especially low‑ and middle‑income ones—to navigate AI governance in a coherent and practical manner. For agrifood systems, this is essential to support cross‑border collaboration on data, early warning, research, and risk management for issues such as food insecurity, animal and plant diseases, and climate shocks. Second, the Dialogue can act as a platform to anchor AI governance in real‑world use cases, ensuring that global norms are informed by sectoral realities. FAO's experience in deploying digital and AI solutions for agriculture—through its Digital Agriculture and AI Innovation Roadmap and AI Governance Toolkit—demonstrates how principles such as human oversight, transparency, inclusion and sustainability can be operationalized in practice for farmers, extension services and public institutions. Third, the Dialogue can strengthen capacity‑building and inclusion, by promoting partnerships across UN entities, governments, academia, civil society and the private sector, and ensuring that the voices of rural communities, women, youth and Indigenous Peoples inform governance outcomes. Finally, by aligning with the Global Digital Compact and broader UN system priorities, the AI Dialogue can catalyze coordinated multilateral action, positioning agrifood systems as a priority sector where responsible AI governance delivers tangible development impact—turning global cooperation into measurable progress on food security and sustainable development.

The AI Dialogue should build upon existing UN system, multilateral, sector‑specific initiatives where AI governance discussions are already advancing, while providing a coherent forum to align them and translate global commitments into practice. At global level, the Dialogue should connect closely with the GDC follow‑up, the UN system‑wide AI governance architecture (including the UN Scientific Panel on AI), and Geneva‑based platforms (WSIS, AI for Good Summit). Strengthened coordination with UN entities active in AI standard‑setting, capacity building and ethics—such as ITU, UNESCO and ODET—can help reduce fragmentation and reinforce interoperability across governance approaches. At the sectoral level, the Dialogue should actively draw on FAO's Digital Agriculture and AI Innovation Roadmap and its AI Governance Toolkit, which offer a concrete, practice‑oriented framework for responsible AI deployment in agrifood systems. These FAO tools are grounded in country experience and address key governance dimensions—risk assessment, human oversight, inclusion‑by‑design, data protection and transparency—relevant to high‑impact AI use cases such as early warning systems, advisory services, disease surveillance and supply‑chain transparency. Existing partnerships with institutions such as CGIAR, the World Bank, regional economic communities, and research and innovation networks, as well as emerging collaborations with Geneva‑based science and innovation actors, also provide ready entry points for scaling responsible AI in agriculture and rural development across regions. The added value of the AI Dialogue lies in its ability to serve as an integrative bridge: linking global AI principles to sector‑specific operational tools, fostering interoperability across initiatives, and catalyzing partnerships that support capacity building, inclusion and country ownership. By explicitly anchoring agrifood systems within international AI governance, the Dialogue can help ensure that global cooperation delivers tangible development outcomes—advancing food security, resilience and sustainability while safeguarding rights and ecosystems.

Governments can contribute by sharing national experiences, regulatory needs, and priority use cases (e.g., early warning, advisory services, disease surveillance) and by committing to implementable follow-up actions for the agrifood sector. Within governments, the statistical authorities are well placed to contribute methodological rigor, data quality standards, and operational experience in producing comparable statistics across very different country contexts. This is relevant to the interoperability and trustworthiness priorities of the Dialogue. UN entities and International Organisations can convene, translate global principles into agrifood guidance, and broker partnerships for capacity-building and institutional readiness. The private sector can contribute technical expertise, risk management practices, and transparency on model/data governance, while supporting responsible access and avoiding vendor lock-in. Academia can provide evidence, benchmarks, and independent evaluation. Civil society, farmers'/producers' organizations/cooperatives and representatives of Indigenous Peoples can ground discussions in lived realities, equity impacts, and accountability needs. For format and structure, the Dialogue could combine: (1) a high-level plenary to agree on objectives; (2) thematic working sessions with clear outputs (recommendations, toolkits, draft principles); (3) sectoral breakouts (including agrifood) focused on concrete use cases and risk controls (e.g., templates for risk/impact assessment and monitoring); and (4) a standing mechanism (e.g., multi-stakeholder working groups) to track implementation between sessions. Clear roles and accountability (including coordination for follow-up) are essential to move from dialogue to implementation. On the practical/logistics aspects, to ensure meaningful and continuous engaged participation, hybrid participation options should be provided, as well as interpretation, accessible materials, and structured ways to submit inputs not only before but also after the Dialogue as follow up to build efficient follow up mechanisms.

Underrepresented perspectives often include smallholder farmers and pastoralists; rural women and youth; Indigenous Peoples and local communities; agricultural extension workers; cooperatives and SMEs; workers across food value chains; consumer organizations; and institutions from low- and middle-income countries with limited resources to engage in global processes. Their participation matters because AI governance choices shape access to information, finance, markets and public services, and can affect land, labor and environmental outcomes. Inclusion can be strengthened by: (1) dedicated seats for farmers' organizations, Indigenous Peoples' representatives, and youth/women networks in panels and working groups; (2) resourcing participation (travel support, connectivity grants, compensated time); (3) multilingual and low-bandwidth engagement channels; (4) co-design of agrifood use cases and safeguards through regional consultations; and (5) a transparent "response-to-inputs" process. Beyond participation, inclusion should be a design requirement for AI systems (participatory design, context testing, usability for low-connectivity and low-literacy settings), with attention to power asymmetries, conflict-of-interest transparency, and accessible grievance channels.

Formats that can drive dynamic engagement include: • Use-case "governance clinics": time-boxed sessions that map a real agrifood AI application (e.g., drought/pest advisory, food insecurity forecasting) and identify risks, controls, and accountability (including risk/impact assessment templates). • Red-teaming and stress-test exercises: participants examine failure modes (bias, data poisoning, model drift, misinformation) and agree on minimum safeguards and monitoring indicators. • Regional listening sessions and field-linked dialogues: hybrid meetings connected to local hubs (extension centers, cooperatives, universities) so rural stakeholders can participate without long-distance travel. • Standards co-creation workshops: hands-on drafting of interoperable templates (impact assessments, data-sharing agreements, model/dataset documentation, and provenance/disclosure practices for AI-generated content). • Marketplace for responsible AI in agrifood systems: curated demonstrations paired with evaluation criteria (safety, inclusion, sustainability) and space for governments to announce pilots or partnerships. To sustain momentum, maintain an online community of practice, an open repository of tools and good practices, and a transparent process to track commitments between sessions.

Examples of good practices and policy approaches that can advance responsible AI governance in agrifood systems include: • Sector-specific AI assurance for high-impact use cases: require risk/impact assessments (proportionate due diligence), human oversight, and validation in local agroecological and linguistic contexts for AI used in advisory services, early warning, food safety, and eligibility/targeting for support programmes. • Data governance frameworks for agrifood data: set rules on purpose limitation, consent, privacy and security; address community and Indigenous data rights and benefit-sharing; and set minimum data quality standards (provenance, metadata, bias checks). • Interoperability standards and shared templates: promote common documentation, data-sharing agreements, and interoperable APIs to enable cross-border collaboration on transboundary pests/diseases, climate risks, and food security early warning while maintaining safeguards. • Digital public goods and open approaches with safeguards: support open-source software, reference datasets and models where appropriate, paired with governance on sensitive data, biosecurity, and downstream accountability; invest in digital public infrastructure countries can adapt and scale; encourage clear intended use/limitations and disclosure where AI-generated content is used. • Capacity-building and institutional readiness: strengthen public institutions, extension services and local innovators through training, guidelines, and "learn-by-doing" pilots; include procurement guidance for responsible AI (transparency, auditability, portability, anti-lock-in). By using the strategic framework and toolkit available in the "FAO's Digital Agriculture and AI Innovation Roadmap", users will be enabled to adopt use of transparent and inclusive criteria for agrifood AI initiatives, such as (i) local readiness (policy, data, infrastructure, trust and skills), (ii) ethics/risk/sustainability assessment including fallback protocols, (iii) accountability and liability arrangements, (iv) grievance and remediation channels, and (v) independent audit expectations and anti lock-in safeguards. • Monitoring and redress: implement continuous monitoring (drift, bias, safety), user feedback loops, and accessible grievance and remedy processes.