Session summaries

Session 1: Exploring the transformative potential of AI for the public sector

Overview

AI was framed not as a narrow technology or efficiency agenda, but as a potential reconfiguration of state capacity. Participants asked whether emerging and middle-income economies can use AI to strengthen health, education, public administration, planning, procurement and citizen engagement, particularly for populations least likely to be served by markets. The core test is whether public sector AI improves access, equity, accountability, trust and the quality of interaction between citizens and the state.

A key distinction was drawn between frontier AI development and effective adoption. Countries do not need to lead in model development to benefit; some may even leapfrog if they can overcome fragmented data, weak infrastructure, skills shortages, procurement bottlenecks and poor institutional coordination. Non-adoption also carries risks as public systems become more overstretched and citizen expectations rise.

Participants identified four foundations for responsible adoption: (1) expanding public value rather than simply cutting costs; (2) building interoperable data systems, connectivity, cloud access, secure data flows and clear stewardship; (3) developing civil servants’ institutional capability to identify, commission, evaluate and govern AI tools; and (4) using sandboxes and readiness-based sequencing to pair quick wins with deeper reforms. While tensions remain around speed, accountability, regulation, sovereignty, partnership, personalisation, equity and democratic purpose, the shared conclusion was that AI adoption should move quickly enough to meet urgent public sector needs, but be governed carefully enough to protect rights, legitimacy and inclusion.

Recommendations

• Anchor AI strategies in public value and inclusion: Define the public purpose of AI before selecting tools, with explicit attention to equity, access, rights, trust and the hardest-to-reach groups.
• Build data foundations before scaling high-risk applications: Prioritise digitisation, interoperability, data quality, metadata, ownership, privacy and trusted data sharing before deploying sensitive AI systems.
• Design accountability into AI systems from the start: Ensure every AI-supported recommendation has a clear human or institutional owner, with audit trails, appeals, explainability, safeguarding and independent evaluation.
• Develop a civil service AI capability model: Build structured AI literacy, implementation capacity and specialist pipelines so officials can commission, use and govern AI responsibly.
• Use sandboxes to manage risk while enabling experimentation: Test AI tools in controlled settings with ethical review, evaluation metrics, user feedback, failure protocols and clear pathways to scale.
• Sequence investments according to readiness: Match implementation to each country’s maturity in infrastructure, governance, skills, cloud access, procurement and priority use cases.
• Communicate AI in practical, citizen-centred terms: Explain AI through tangible benefits such as shorter waits, better access, earlier interventions, reduced paperwork and improved learning support.
• Treat participatory governance as a source of legitimacy: Involve civil servants, technical experts, frontline workers, regulators, private sector, civil society and affected communities throughout design and deployment.
• Measure what matters: Track service quality, inclusion, frontline time saved, error reduction, fraud detection, citizen satisfaction, trust, rights impacts and distributional effects, not just deployment or efficiency.
• Frame the next phase around AI readiness: Move from potential to implementation by mapping countries or institutions against data, infrastructure, governance, talent, trust, procurement, fiscal capacity and use-case priorities.

Session 2: Assessing AI readiness and using AI frameworks

Overview

A central question was how governments, especially in emerging economies, can assess AI readiness and move from experimentation to responsible implementation. Public sector adoption is already underway through drafting, analysis, service support, productivity tools, strategies, training, regulation and coordination efforts, though maturity varies widely. The challenge is to turn this momentum into trusted, inclusive and measurable public value rather than a patchwork of fragmented pilots.

AI was framed as a state capability agenda shaped by fiscal pressure, rising citizen expectations, uneven infrastructure and competition for skills, investment and influence. Participants saw readiness as requiring leadership, coordination, data governance, infrastructure, skills, financing, procurement capacity and public trust. They emphasised shifting from activity metrics, such as pilots launched or officials trained, to outcomes such as faster processing, fiscal savings, fraud reduction, better targeting, improved service quality, citizen satisfaction and equitable access across areas including audit, tax, welfare, agriculture, health, education, procurement and citizen services.

Human capacity, institutional culture, inclusion and local relevance were treated as core conditions for responsible adoption. Leaders, legislators, technical teams, frontline staff and users need role-specific AI literacy, while governments need clear rules, approved tools, sandboxes and human oversight to avoid both paralysis and unsafe use. AI strategies must account for low connectivity, local languages, varied literacy, disability access, human-in-the-loop models and offline alternatives, adapting global frameworks to local realities. The shared conclusion was pragmatic: move quickly, and intentionally, enabling safe use now while building long-term foundations in law, data, infrastructure, skills, research capacity, public communication and institutional learning.

Recommendations

• Establish a whole-of-government AI coordination mechanism: Create or strengthen a central function to set standards, convene ministries, coordinate procurement, manage risk, track implementation and maintain a national use-case portfolio while allowing departmental innovation.
• Conduct a practical AI readiness assessment: Assess institutional capacity, infrastructure, data systems, skills and policy maturity, potentially using the PwC archetypes as a starting point for sequencing feasible AI pathways.
• Move from pilots to a prioritised national use-case portfolio: Build a structured use-case library and prioritise projects by feasibility, public value, risk and capacity, with early pilots delivering measurable results within 6–12 months.
• Define impact metrics before implementation: Set success measures upfront, including processing time, savings, fraud detection, accessibility, complaints, satisfaction, staff time saved, backlog reduction and equity of access.
• Build role-specific AI skills across the public sector: Tailor training for leaders, managers, technical teams and frontline staff, covering strategy, workflow redesign, data, cybersecurity, monitoring, safe use and citizen interaction.
• Create safe AI environments for public servants: Provide approved tools and secure experimentation spaces, supported by data classification rules, model-use protocols, human review and error reporting channels.
• Invest in inclusive, locally relevant AI systems: Design for local languages, low bandwidth access, mobile and SMS channels, varied literacy levels, disability access, human oversight and offline options for critical services.
• Strengthen legislative and public communication capacity: Equip legislators, regulators and communicators to scrutinise AI, explain safeguards, address risks and build public confidence without hype or fear.
• Use regional cooperation to close capability gaps: Develop shared guidelines, training, research partnerships, pooled expertise and joint infrastructure, including for language models, data governance and AI ethics.
• Treat AI as a long-term state capability agenda: Build institutions able to learn and adapt over time, linking AI to fiscal policy, education, public administration reform, infrastructure, research, competitiveness and trust.

Fireside chat with Google DeepMind

A central discussion at the event explored the critical transition of AI from cutting-edge science to delivering tangible real-world impact, including within government. The conversation focused on how leaders can navigate the opportunities, challenges, and key questions associated with AI adoption.

The following core ideas emerged from the session:

• Purpose-driven AI: A fundamental premise was that the pursuit of AI should not be about technological supremacy, but rather focused on solving humanity’s most pressing challenges whilst mitigating risks and maximising benefits.
• Inclusive, community-led adoption: To ensure technology bridges gaps rather than exacerbating them, society must be brought along as an active partner. It was strongly advised that AI initiatives should be built with communities, rather than to them, ensuring local voices help steer the transition from the very beginning.
• Responsible pioneering and safety by design: Overcoming institutional resistance requires a mindset of continuous improvement, or ‘kaizen’, which involves the humility to recognise broken systems and the optimism to mend them. Crucially, responsibility must be built into systems, not bolted on as an afterthought. Just as safety is foundational when building aeroplanes or bridges, robust AI guardrails act much like brakes on a vehicle; they do not slow progress, but rather provide the confidence required to move swiftly and safely.
• Strategic implementation (‘Start Small, Think Big’): To transition from the ‘why’ to the ‘how’ of AI adoption, leaders were encouraged to adopt a practical approach by targeting low risk, high return use cases first. This strategy allows governments to prove the technology’s value before scaling up.
• Human capital and co-creation: Technology is only meaningful if it improves people’s lives, which necessitates a focus on building human capital rather than merely deploying digital tools. Successful national partnerships must be rooted in co-creation rather than simply delivering ‘black box’ solutions. By upskilling the local workforce, nations ensure their citizens are empowered with the agency to own, build, and steer their own futures, rather than remaining mere consumers of technology.
• Impact of frontier AI: The discussion highlighted breakthroughs such as AlphaFold, which represents a landmark advance in solving the decades old protein-folding problem by predicting 3D structures from amino acid sequences. Furthermore, by providing an open access database of over 200 million protein structures, this technology has reshaped drug discovery and agricultural resilience, whilst accelerating global efforts to combat challenges like malaria and antimicrobial resistance.

Session 3: Maximising potential, mitigating risks, and ensuring responsible governance

“AI adoption succeeds or fails on institutional capacity — on governments’ ability to define public value, build trust, empower leaders, equip people, learn by doing, and align technology with real-world service delivery.”

Overview

Experts discussed how governments can maximise AI’s developmental and public sector benefits while managing risks through enforceable governance, institutional accountability and inclusive design. Participants agreed that AI is becoming core social, economic and political infrastructure, but that governance is lagging behind deployment, especially in public administration. Evidence from a forthcoming global study (unreferenced) highlighted the gap: across 135 countries, the average responsible AI score was 35/100; only 55% of nearly 500 assessed frameworks were under implementation; and public sector AI governance was the weakest area.

Trust was framed as a product of accountability, not communications. Because government use of AI is an exercise of state power, participants argued it should meet a higher standard than private sector experimentation. Concerns included opaque deployment, weak remedies for affected citizens, discrimination, mass surveillance, social scoring, misinformation, technology-facilitated gender-based violence and rights violations in lower capacity governance environments. Regulation was seen as a way to create predictable obligations, reduce legal uncertainty and avoid leaving disputes to fragmented court processes.

Participants favoured coordinated but distributed governance, with sector regulators managing domain-specific risks and a central body ensuring coherence, filling gaps and supporting cross-sector learning. Inclusion and digital sovereignty were recurring themes: AI systems must be designed to address disability access, low-resource languages, communities absent from dominant datasets and unequal procurement choices. Sovereignty was defined less as self-sufficiency than as agency, resilience, choice, bargaining power and the ability to shape national digital futures. Across discussions around speed, regulation, sovereignty, market power, geopolitics and growth, the central message was that AI policy should start with public interest goals and judge AI by whether it measurably, equitably and accountably improves citizens’ lives.

Recommendations

• Build enforceable governance foundations, not only principles: Create binding rules for high-risk AI, especially in public administration, covering procurement, disclosure, audits, contestability, redress and safeguards against discriminatory or coercive use.
• Hold government AI-use to a higher standard: Require public bodies to disclose where AI is used, what it does, what data it relies on, how decisions can be challenged and what remedies exist.
• Make redress central to responsible AI: Ensure accessible appeal, complaint, review and remedy mechanisms where AI affects rights-bearing services such as benefits, healthcare, education, enforcement, immigration or taxation.
• Develop coordinated but distributed regulation: Combine sector-specific oversight with a central coordinating body that ensures coherence, closes gaps and supports cross-sector learning.
• Fund independent public interest capacity: Resource civil society, academia, public interest technologists, consumer advocates, disability organisations and affected communities so participation is substantive.
• Prioritise inclusion from design: Require inclusive datasets, local language performance, accessibility testing, disability inclusion and evaluation across population groups in design, procurement and funding.
• Strengthen AI literacy and democratic participation: Invest in public AI literacy and civic mechanisms that help citizens understand, influence and contest AI-enabled decisions.
• Use procurement and disclosure to shape markets: Require interoperability, transparency, accessibility, data protection, contestability and protections against vendor lock-in; encourage AI disclosures, workforce impacts and risk controls.
• Pursue pragmatic digital sovereignty: Treat sovereignty as agency, resilience and choice through multi-cloud strategies, interoperability, open-source options, local model adaptation, domestic skills, regional infrastructure and stronger bargaining capacity.
• Link national regulation with international coordination: Promote interoperable frameworks, shared minimum standards, support for emerging economies and safeguards against a race to the bottom.
• Address energy, compute and infrastructure constraints honestly: Align AI strategies with electricity, data centre capacity, connectivity, compute access and sustainability, avoiding speculative investments that divert resources from core development needs.
• Define the North Star before choosing technology: Start with concrete public interest goals, then shape AI policy, infrastructure, regulation and partnerships around measurable improvements in citizens’ lives.

Session 4: Developing human capital strategies

Overview

Public sector AI transformation is an institutional and human-capital challenge, not simply a technology or procurement exercise. AI is already in government, often through informal ‘shadow’ use. Governments now see the opportunity to shift from passive adoption to active stewardship in the public interest. Success requires defining public value before scaling activity: better services, stronger state capability, inclusion, trust, accountability and productivity, rather than counting licences, pilots, or training.

Participants emphasised that established public service culture, norms and incentives are relevant barriers. Public servants often lack permission to experiment, share data or approve deployment because saying “no” is safer than saying “yes”. Therefore,  leadership appears as a critical point in AI adoption journeys: leaders must model responsible AI use, protect human judgement, reward learning and create conditions for experimentation that is safe, governed and purposeful. Skills strategies must also go beyond technical training, using capability audits, foresight, differentiated learning pathways and multidisciplinary teams that include engineers, lawyers, ethicists, domain experts, public servants and communities.

Local delivery, shared infrastructure and comparative advantage were recurring themes. Municipalities are closest to many AI-relevant services but are often underrepresented in national strategies; shared platforms and reusable tools can reduce duplication if they preserve local adaptation and trust. Participants cautioned that most countries should focus less on competing across the full AI stack and more on areas of distinctive value, such as language, culture, agriculture, energy, health, education and service delivery. Across tensions around sovereignty, speed, safeguards, standardisation, workforce needs, leadership and long-term value, the central conclusion was that AI transformation will succeed or fail on institutional capacity: the ability to define public value, build trust, equip people, learn responsibly and align technology with real service-delivery needs.

Recommendations

• Define success before scaling activity: Align Kepis with public outcomes such as service quality, capability, inclusion, trust, productivity and accountability, not just procurement, usage or training volume.
• Build human capital strategies around institutional conditions: Pair skills programmes with reforms to incentives, leadership, governance, data access and accountability.
• Reward responsible approval and experimentation: Reduce the career risk of saying “yes” by rewarding responsible initiative, documented learning and evidence-based adaptation.
• Use problem-led deployment: Apply AI to concrete service problems, such as hospital waits, benefits eligibility, teacher support, correspondence, complaints handling and local revenue administration.
• Create multidisciplinary AI teams: Combine technical, legal, ethical, procurement, domain, frontline and community expertise from design through testing.
• Develop differentiated skills pathways: Distinguish baseline literacy, role-specific capability, advanced technical expertise, leadership skills and citizen-facing inclusion.
• Embed AI capacity inside ministries and local administrations: Place AI and data professionals within delivery institutions so capability grows inside operational workflows.
• Strengthen local government participation: Involve municipalities through peer networks, coalitions, shared platforms and locally adapted pilots.
• Invest in shared infrastructure where functions are common: Build common compute, data banks, marketplaces, templates, governance tools and reusable service modules while allowing local adaptation.
• Focus national capability on comparative advantage: Prioritise areas of distinctive public value rather than attempting to compete across the entire AI stack.
• Treat trust as an implementation requirement: Design for accountability, transparency, confidentiality, ethics, safety and fitness for purpose, with user co-design where trust is fragile.
• Document and share culture change examples: Create a practical repository of leadership behaviours, incentive changes, governance models, pilot structures and lessons from success and failure.
• Balance rapid deployment with continuous monitoring: Move without waiting for perfect frameworks, but include guardrails, evaluation, safety review and revision mechanisms.
• Prepare for labour market transition: Link reskilling for affected workers to sectors with future demand and social need, not only to AI jobs.

Session 5: What are the policy enablers critical for AI success?

Overview

Participants agreed that AI is not merely a technology to buy, but a public sector transformation agenda requiring political leadership, adaptive regulation, infrastructure, financing, institutional capability and public legitimacy. The core challenge is to deliver visible public value quickly while avoiding fragmented pilots, vendor dependency, weak safeguards and shallow ‘innovation theatre’.

AI policy was framed as a cross-government priority linked to industrial policy, education, fiscal policy, procurement, public administration, foreign policy and sovereignty, especially for emerging economies facing constrained capacity, limited compute, external financing dependence, linguistic exclusion and externally designed systems. Trust must be earned through tangible benefits, credible safeguards, transparency, inclusion and institutional performance. Participants called for senior political sponsorship and central coordination, paired with deployment through line ministries in sectors such as health, education, agriculture, transport and housing, using real service-delivery problems as the basis for learning.

Pragmatic, adaptive and risk-based regulation was seen as essential where AI affects rights, services, safety or accountability, with human oversight, transparency, validation, auditability and redress as legitimacy conditions. Data, compute, interoperability and local-language resources were treated as foundational infrastructure, while sovereign AI was defined as policy agency over standards, procurement, data governance, model behaviour, contracts and strategic choices rather than simple hardware ownership. Participants strongly rejected “pilot-itis”: AI initiatives should be designed from the outset for validation, procurement, financing, deployment, monitoring and institutional ownership. Regarding speed, sovereignty, co-design, evidence and dependency, the shared conclusion was that governments should act quickly but govern adaptively, protect the decision layer, build trust through results and safeguards, and treat AI as part of wider public sector reform.

Recommendations

• Define success before scaling activity: Align KPIs with service quality, institutional capability, inclusion, trust, productivity and accountability, not just procurement, training or usage.
• Establish AI as a strategic government priority with senior sponsorship: Assign responsibility at the centre, empower coordination and tie senior review to delivery.
• Adopt adaptive, risk-based regulation: Use principles-based frameworks, co-designed with stakeholders and revised periodically, with stronger safeguards for high-impact uses.
• Build trust through visible public value: Prioritise citizen-facing use cases in areas such as health, agriculture, education, traffic and service delivery.
• Require human oversight, transparency, validation and redress in high-impact systems: Define oversight requirements, publish accessible explanations, validate systems and maintain audit and complaint mechanisms.
• Treat data and local language resources as public infrastructure: Invest in interoperable registries, machine-readable data, data-sharing frameworks and inclusive local-language datasets.
• Pursue hybrid infrastructure and protect the decision layer: Use cloud where useful while retaining control over standards, procurement, data governance, contracts and exit options.
• Use procurement as a market shaping tool: Require interoperability, transparency, clear data rights, exit plans and avoidance of vendor lock-in while supporting local capability.
• Move beyond pilots by designing for scale from the outset: Link experimentation to validation, procurement, financing, deployment, monitoring and public sector ownership.
• Invest in the civil service middle layer: Build practical AI capability among commissioners, procurement officials, regulators, programme managers, evaluators and sector administrators.
• Finance institutions and people, not only projects: Align AI strategies with domestic budgets and use external funding to strengthen national priorities, talent and long-term capability.
• Negotiate partnerships on sovereign terms: Scrutinise data rights, model-training terms, hosting, intellectual property, continuity risks and exit conditions; use regional cooperation to improve bargaining power.
• Create safe learning and reporting ecosystems: Encourage reporting of failures and near misses through non-punitive systems, while preserving accountability for negligence or avoidable harm.
• Embed inclusion from the design stage: Treat gender, language, skills and access gaps as core design constraints, not add-ons.
• Use scenario planning for macro-level disruption: Prepare for futures where AI raises productivity while disrupting employment, labour markets and social protection.
• Frame AI policy as public sector transformation: Connect AI adoption to budgeting, procurement, institutional design, service delivery, accountability and trust.

Sessions 6 and 7: Generating actionable roadmaps

Overview

Breakout sessions translated shared learning into practical roadmaps for AI-enabled government transformation in emerging economies. Across the four groups, participants stressed that AI should not be treated as an end in itself, but as a tool for solving public problems, improving services and delivering national, regional and local missions.

Several common themes emerged. First, governments need clearer shared language, practical cooperation mechanisms and trusted spaces to align on terminology, red lines, standards and implementation challenges. Second, emerging economies face major infrastructure and investment barriers, including limited compute, data centres, subsea connectivity and interoperable public data systems. Regional collaboration was seen as essential to create scale, attract investment and support shared assets such as compute networks, data exchanges and reusable code or policy resources.

Participants also emphasised that responsible AI adoption depends on visible senior leadership, civil service capacity, public trust and inclusion. This includes targeted training for officials, incentives for institutional reform, stronger university and regulator partnerships, and attention to exclusion risks such as language bias. Public communication should focus less on AI as a technology and more on tangible benefits, mission-led problem solving and trust-building. Practical tools, such as problem-statement blueprints, trust playbooks, shared standards, public perception surveys and cross-country learning platforms, were identified as ways to move from principles to implementation.

Recommendations

• Create nimble policymaker alignment groups: Convene small, practical groups to clarify terminology, identify constraints and red lines, and turn broad AI governance principles into implementable cooperation.
• Strengthen peer-to-peer implementation exchanges: Support direct learning among governments, regulators and standards bodies on domestic public sector AI use, procurement, portability, switching, bias testing and public sector protocols.
• Use universities as trusted capacity partners: Engage universities to help governments understand AI policy, train public servants and support implementation in ways that are credible and locally grounded.
• Run public perception surveys on AI: Build on examples such as South Africa to assess public attitudes across regions, including Latin America and Africa, and use the findings to guide trust-building and communication.
• Pool regional demand for AI infrastructure: Use regional scale, shared compute, cross-border data embassies and supercomputer networks to attract investment in GPUs, data centres and connectivity.
• Reframe AI sovereignty around control and agency: Define sovereignty as the ability to protect citizens’ data and direct national technological development, rather than relying only on strict data localisation.
• Build interoperable, AI-ready public data systems: Move beyond fragmented ministry-level systems by strengthening central digital public infrastructure, data exchanges and coordination across government.
• Scale inclusive AI literacy and civil service training: Provide broad public digital literacy, targeted training for officials and specific action on language bias, including in underrepresented linguistic communities.
• Incentivise civil service upskilling and reform: Use competitive frameworks, awards and visible success stories to motivate institutions, build trust and improve digital service delivery.
• Secure senior leadership commitment: Use Cabinet-level AI sensitisation, executive awards and formal safety standards to generate political buy-in, budgets and sustained attention.
• Establish a shared knowledge and code hub: Create a cross-country platform for reusable code, prompts, research, policy materials, use cases, failures and communications lessons to reduce duplication.
• Support SMB adoption through Digital Public Infrastructure (DPI) and partnerships: Provide SMB toolkits, cloud credits, workshops and partnerships with local technology firms and business chambers to expand practical AI adoption.
• Develop harmonised ‘train the trainers’ programmes: Use specialised curricula, standard assessments and tiered facilitators to build public sector capability, including in areas such as audit and accounting.
• Shift public communication from technology to benefits: Use clearer language, stronger visuals and shared narratives focused on everyday outcomes rather than promoting AI as the central message.
• Create a mission-led Problem Statement Blueprint: Help governments and sectors define shared challenge statements linked to funding, deployment, inclusion and opportunities for scaling proven solutions.
• Develop a practical Trust Playbook: Provide scenario-based guidance for government and innovation actors on how to enact, demonstrate and communicate trust in local, national and regional contexts.

Session 8: Feedback on actionable roadmaps

“AI adoption should not be pursued for its own sake, but as a means to improve service delivery, strengthen social outcomes, increase administrative efficiency, and enable locally relevant innovation.”

Overview

There was discussion on how to turn broad AI principles into practical roadmaps for governments and partner institutions. Participants agreed that AI should not be adopted for its own sake, but used to improve public services, strengthen social outcomes, increase administrative efficiency and support locally relevant innovation. Because governments vary widely in institutional capacity, infrastructure, regulatory maturity, procurement capability, public trust, compute access, standards expertise and university capacity, participants favoured peer learning, shared vocabulary, practical toolkits and adaptable cooperation models over a single universal approach.

Trust was a central theme, especially the distinction between abstract trust in AI and citizen trust in government use of AI in specific services such as health, education, welfare, justice, identity, public safety and administration. Participants argued that trust should be built through visible, mission-oriented examples of public value, not technology-led messaging. Governments also need crisis-response protocols for failures involving bias, privacy breaches, data loss, service disruption or harmful automated decisions.

Standards, evaluation, procurement, evidence, infrastructure and local innovation were treated as core enablers of responsible adoption. Shared approaches to bias testing, interoperability, portability, switching and procurement safeguards could reduce fragmentation and vendor dependency, while an evidence hub could capture case studies, procurement lessons, standards guidance, infrastructure models and public engagement examples. Participants also emphasised compute, data systems, digital identity, interoperable platforms and regional pooling options such as shared compute, supercomputing networks and data embassies, alongside support for local SMEs developing tools suited to local languages and service needs. Varying views included whether to foreground AI or service outcomes, how to measure trust, how much to centralise implementation, how to balance local adaptation with regional pooling, and how to encourage experimentation while preserving accountability.

Recommendations

• Create practical peer-learning groups. Form nimble cross-country groups for policymakers, regulators, standards bodies, and implementers to clarify shared challenges, define red lines, compare policy objectives, and build common language on AI adoption, infrastructure, standards, procurement, and trust.
• Build public sector AI capability and leadership. Use peer exchanges, civil service education, train-the-trainer models, and targeted AI literacy to strengthen senior leadership and implementation capacity across procurement, oversight, interoperability, safety testing, and digital standards.
• Develop university-led AI diplomacy and capacity building. Use universities as trusted conveners for AI diplomacy programmes, executive education, retreats, and alumni networks that connect senior civil servants, regulators, standards bodies, and technology-sector participants.
• Create a trust playbook for government innovation. Develop guidance for understanding citizen concerns, earning trust, communicating transparently, and responding credibly when AI public services fail or cause harm.
• Use public perception surveys to measure trust in context. Deploy or expand surveys that assess citizen trust in government use of AI by service area, including comfort with specific applications and concerns about privacy, employment, data misuse, bias, and exclusion.
• Lead with public outcomes, not AI branding. Frame adoption around visible improvements in health, education, welfare, justice, administration, and public safety rather than abstract claims about AI transformation.
• Strengthen standards, evaluation, and procurement safeguards. Cooperate on bias testing, interoperability, portability, switching, safety evaluation, and safeguards against unsuitable or poor-quality systems, adapting existing open-source evaluation approaches where useful.
• Build a shared evidence and knowledge hub. Collect cross-country examples of successful and unsuccessful public sector AI adoption, including case studies, procurement lessons, infrastructure models, standards guidance, and public engagement practices.
• Use mission-oriented and challenge-based procurement. Define clear public problems and invite solutions, using challenge models to stimulate innovation, demonstration, and public understanding.
• Pool infrastructure where national capacity is limited. Explore shared compute, regional supercomputing, data embassies, and collective investment strategies while protecting sovereignty, data control, and public value.
• Develop interoperable digital public infrastructure
  Build reusable government platforms and components, such as notification systems, data exchange tools, identity systems, and service-delivery modules, rather than duplicative ministry-by-ministry systems.
• Support local innovation ecosystems. Use procurement, digital public infrastructure, language needs, and local business support to help SMEs and local developers create AI tools suited to national and regional priorities.
• Align incentives with responsible adoption. Design awards, rankings, evaluations, and recognition systems carefully so they encourage useful innovation and better service delivery without promoting box-ticking, reputational avoidance, or risk-shifting.

Session 9: Forging partnerships and accelerating actionable initiatives

Overview

In this session, leaders examined how partnerships can move AI governance and deployment from aspiration to population-scale impact. Participants framed AI as critical infrastructure with major implications for sovereignty, trust, competitiveness and state capacity. For emerging economies, the challenge is not simply to adopt AI, but to shape what is built, whose data is used, who governs it and whose interests it serves.

Trust was treated as a design requirement, not a communications problem. Citizens need to see that AI systems are competent, accountable, rights-respecting and aligned with public interest, particularly amid concerns about misinformation, platform harms, corporate concentration and weakened public confidence. Participants stressed that national AI strategies are insufficient without execution capacity, including broad AI literacy across policymakers, regulators, procurement officials, educators, health administrators and frontline public servants. Partnerships should therefore transfer capability, not merely deliver products.

AI was also discussed as shared infrastructure, including digital public infrastructure, interoperable data systems, national and regional data spaces, federated networks, sovereign compute and reusable public interest tools. Smaller, cheaper and locally deployable models may offer a more sovereign, affordable and energy-efficient path than reliance on frontier models alone, especially when paired with local languages, domestic scientific datasets and sector specific use cases in health, education, agriculture and climate resilience. Participants warned against the ‘pilot trap’, arguing that initiatives need evidence, financing, accountability, political ownership, clear success metrics and kill switches from the outset. Across tensions around speed, trust, autonomy, scale, stewardship and political cycles, the strongest partnerships were seen as those that give emerging economies agenda-setting power, pool regional influence and anchor AI initiatives beyond individual administrations or funding windows.

Recommendations

• Make trust a design requirement, not a communications objective: Build systems that are useful, transparent, accountable and visibly aligned with citizens’ interests.
• Build partnerships in which emerging economies are agenda setters: Ensure governments, civil society, researchers, communities and local innovators define priorities, use cases, data governance and infrastructure choices.
• Invest in collective regional and middle-power coalitions: Pool markets, compute, standards, data governance and negotiating power through regional alliances and shared infrastructure.
• Treat AI as critical public infrastructure: Prioritise digital public infrastructure, interoperable systems, national data spaces, federated networks and shared compute.
• Create applied mission labs for public interest AI: Establish local, mission-driven labs focused on health, education, agriculture, climate resilience, social protection and public administration.
• Develop sovereign AI through smaller and localised models: Use domestic scientific, linguistic and sectoral data to build affordable, efficient and governable models.
• Embed AI capability across the public sector: Train and place AI-literate talent across ministries, regulators, procurement teams and frontline services.
• Reform procurement toward co-creation and capability transfer: Move from vendor purchasing to co-development, embedded teams and long-term public sector capability building.
• Use blended finance and outcome-based contracting carefully: De-risk public-interest AI while tying finance to measurable social outcomes and avoiding distorted priorities.
• Fund evaluation as a core project component: Include evaluation budgets, success metrics, staged evidence models and kill switches from the start.
• Design for scale, sustainability or termination from the start: Give every pilot a clear path to public-system scale, local sustainability or closure.
• Strengthen regulatory sandboxes and data trusts: Test AI under oversight and use trusted institutions to manage data access, ethics and accountability.
• Develop AI crisis playbooks: Plan for failures, withdrawals, systemic harms and service disruptions through red-teaming, recovery plans and interdependency mapping.
• Promote shared tools, open-source assets and reusable infrastructure: Reduce duplication through common protocols, reusable components and adaptable public sector AI tools.
• Anchor partnerships beyond political cycles: Use independent institutions, statutory mandates, multi-year financing and cross-party ownership to sustain initiatives.
• Use convening as a pathway to collective action: Turn dialogue into coalitions, shared infrastructure proposals and implementable partnership models.