IGF 2025 WS #294 AI Sandboxes: Responsible Innovation in Developing Countries

The “AI Sandboxes: Responsible Innovation in Developing Countries” workshop convened during IGF 2025 explored sandboxes as tools for AI governance. The session brought together a diverse group of stakeholders from governments, civil society, the private sector, and technical communities. The goal was to explore how AI sandboxes can be leveraged to promote innovation while ensuring responsible and inclusive governance, particularly in the Global South. The discussion revolved around the potential of sandboxes, challenges to implementation, best practices, and cross-border cooperation, with a focus on sectors such as health.

Sophie Tomlinson, Director of Programs at the Datasphere Initiative, opened the session by sharing some of the governance challenges posed by rapid AI integration. She introduced a Mentimeter to engage attendees and invited them to share their views and knowledge of sandboxes. Mariana Rozo-Paz, Research and Project Management Lead at the Datasphere Initiative, provided a conceptual and practical overview of sandboxes. She described them as collaborative, safe spaces where stakeholders such as regulators, the private sector, and civil society can experiment with technologies against existing or evolving regulatory frameworks. Rozo-Paz differentiated between regulatory, operational, and hybrid sandboxes, noting their roots in fintech and their expansion into domains like health and transportation.

Meni Anastasiadou, Digital Policy Manager at the International Chamber of Commerce (ICC), highlighted the role of AI in transforming industries and improving productivity. She shared how small and medium enterprises (SMEs), are often unequipped to handle fragmented regulatory environments. Anastasiadou presented ICC’s Four-Pillar Governance Framework for AI: harmonization with existing international agreements, simplification of compliance, regulatory flexibility to attract investment, and fostering trust. She highlighted the value of sandboxes as a space where all stakeholders can safely test and iterate, likening them to engineers testing planes before flight.

Alex Moltzau, Policy Officer at the European AI Office, reflected on his experience working on privacy sandboxes in Norway and now helping implement AI sandboxes across the EU. He discussed the European Commission’s current efforts to coordinate regulatory sandboxes via the AI Act, with a particular focus on ensuring free access for SMEs. Moltzau stressed that sandboxes can help provide evidence-based policy development and that learning through experimentation is crucial for responsible AI innovation. He also pointed to emerging plans for cross-border sandboxes under the Interoperable Europe Act and encouraged global stakeholders to provide feedback on upcoming EU consultations.

Jimson Olufuye, Chair of the Advisory Council at AFICTA, offered a regional perspective from Africa, particularly Nigeria. He emphasized that fewer than 10 African countries have developed national AI strategies. Sandboxes, he argued, could accelerate both regulatory learning and product development. Recalling a recent co-creation lab in Abuja organized with the Datasphere Initiative, he noted that Nigeria’s regulators, including the Nigerian Data Protection Commission and the Communications Commission, are beginning to adopt sandbox methodologies in data protection and fintech.

Thiago Guimarães Moraes, a researcher at Vrije Universiteit Brussel, emphasized the importance of piloting through what he referred to as “sandboxing the sandbox.” He recommended that regulators start small, leveraging internal expertise and international partnerships to manage resource constraints. Moraes also called for greater civil society involvement, noting that individuals affected by AI systems must be represented before, during, and after sandbox operations.

Morine Amutorine, Africa Sandboxes Forum Lead at the Datasphere Initiative, expanded on regional barriers in Africa. Of the 25 national sandboxes mapped across the continent, 24 are in finance. Regulators are enthusiastic but often lack clarity on scope, legal authority, and funding models. Amutorine outlined Datasphere’s support programs and work on sandboxes, including co-creation labs, technical coaching, and exploration of shared-cost funding models across regulatory bodies.2 She stressed the importance of identifying clear legal mandates and sectoral engagement early in the sandbox process.

Jai Ganesh Udayasankaran, Executive Director at AeHIN, provided insights from the health sector in South and South-East Asia. He described three key use cases for sandboxes in health: bringing private sector applications into public Universal Health Coverage (UHC), validating interoperability with national health data standards, and managing responsible cross-border health data sharing.3 He also noted the diversity of terminology across the region such as regulatory sandbox, testbed, and living lab, and emphasized the need for convergence and harmonization. He opined that apart from the traditional role of sandboxes being regulator-owned gatekeeping mechanism they could function as collaborative environment where innovative solutions from new and emerging technologies can be co-created in coordination and cooperation with the relevant stakeholders.

Natalie Cohen, Head of the Regulatory Policy for Global Challenges unit at the OECD introduced the OECD’s 2021 Recommendation on Agile Regulatory Governance and stressed that while sandboxes are valuable tools, they are not suitable for all contexts. Policymakers must clearly define goals, eligibility criteria, evaluation metrics, and exit strategies. Cohen noted that sandboxes are most effective when they support early commercialization and regulatory learning, but they require significant funding, staffing, and legal clarity. She also warned of potential market distortions if sandboxes inadvertently favor large companies over SMEs, urging governments to consider financial support and data access for smaller players.

The workshop confirmed that AI sandboxes are rapidly evolving into critical tools for AI governance, particularly in developing countries seeking to balance innovation with accountability. While enthusiasm is growing across regions and sectors, participants highlighted the need for legal clarity, inclusive participation, transparent evaluation, and sustainable funding models. Cross-border collaboration and trust-building were underscored as essential for sandboxes to truly fulfill their promise. As global frameworks like the EU AI Act move forward, work on sandbox capacity building will remain vital to ensure that sandboxes are not only technically sound but also socially legitimate and responsive to local contexts.

1 See, e.g., Datasphere Initiative (2025). Sandboxes for AI: Tools for a new frontier. https://www.thedatasphere.org/datasphere-publish/sandboxes-for-ai/

2 See Datasphere Initiative (2025). Africa Sandboxes Outlook: Thinking outside the box in the age of AI.
https://www.thedatasphere.org/datasphere-publish/africa-sandboxes-outlo…

3 See, e.g., Asia eHealth Information Network (26 February 2025).
Exploring sandboxes in the health sector: Why, when, and how to support responsible innovation?
https://youtu.be/bmY2ZVCf_KI?. See also Digital Health Convergence (2025). Digital Health Convergence
Workshops. https://digitalhealthconvergence.net/convergence-workshops &
Digital Health Convergence (2025). GAPS 2.0 A call to action towards strengthening digital health ecosystem in Asia.
https://digitalhealthconvergence.net/gaps-framework?section=2