India AI Impact Summit 2026

ai-dpi-experimentation-and-the-role-of-sandboxes

7 min read

TRACK:Safe and Trusted AI
SPEAKERS
  • Adesh Khadka, Ministry of Communication and Information Technology- Government of Nepal
  • Adesh Khadka, Ministry of Communication and Information Technology- Government of Nepal
  • Alexandru Oprunenco, UNDP Asia-Pacific
  • Dr. Nkundwe Moses Mwasaga, Information And Communication
  • Dr. Verena Kontschieder, Opendata.ch; Prototype Fund Switzerland
  • Kavita Bhatia, Ministry of Electronics & IT, GoI
  • Lorrayne Porciuncula, Executive Director
  • Morine Amutorine, Datasphere Initiative
  • Sushant Kumar, Kalpa Impact
SESSION OVERVIEW

This session explores how sandboxes can support safe, accountable, and trustworthy AI adoption in DPI systems. Through presentations and panel discussion, participants will engage with practitioners to examine challenges and complexities introduced by AI-enabled capabilities in DPI. The conversation will focus on how sandboxes provide controlled settings to test new technologies while fostering trust through diverse stakeholder involvement. Participants will leave with practical governance insights and design principles for building inclusive and scalable AI-powered DPI.

VIDEO RECORDING

AI-DPI Experimentation and the Role of Sandboxes

Abstract

The session examined how sandbox environments can enable safe, accountable, and trustworthy deployment of AI within large‑scale Digital Public Infrastructure (DPI). After a framing keynote by India’s DPI architect, Kavita Bhatia, the DataSphere Initiative unveiled its first global report on DPI sandboxes. A moderated panel then explored practical experiences from Tanzania, Nepal, UNDP, Africa, and Switzerland, discussing challenges, governance mechanisms, and strategic choices for AI‑enabled DPI. The dialogue emphasized the need for contextual experimentation, inclusive design, and institutional learning to build citizen trust at population scale.

Detailed Summary

  • Moderator Sushant Kumar opened the session with a quick poll on familiarity with sandboxing, DPI, and AI, confirming a highly knowledgeable audience.
  • He positioned digital public infrastructure (DPI) as the essential layer for public‑service delivery, now being overlaid with AI capabilities.
  • Emphasised the “speed of trust” thesis: without coordinated trust among government, civil society, and the private sector, AI‑enhanced DPI risks stalling or failing.
  • Outlined the session’s agenda: a keynote, a short video/report launch, and a panel discussion focused on experimentation through regulatory, policy, and technology sandboxes.

2. Keynote – “Sandboxing at Population Scale” (Kavita Bhatia)

2.1 India’s DPI Journey

  • Described India’s India Stack (Aadhaar, Unified Payments Interface, Digital Payments, etc.) as a proof‑of‑concept that inclusive, interoperable, and innovation‑friendly DPI can be built for 1.4 billion citizens.
  • Highlighted measurable outcomes: mass adoption of QR‑code payments, increased financial inclusion, and heightened transparency that reinforced citizen trust.

2.2 AI as a New Capability Layer

  • Defined AI as predictive intelligence, adaptive automation, and large‑scale decision support that can shift government from reactive to proactive service delivery.
  • Cautioned that AI introduces systemic bias, opacity, data‑governance risks, and accountability challenges that scale as quickly as the benefits.

2.3 Why Experimentation Is a Governance Necessity

  • Presented three core sandbox functions for India:

    1. Controlled environment – real data testing to observe actual algorithmic outcomes.
    2. Risk anticipation & mitigation – early detection of bias or unfairness before full rollout.
    3. Iterative governance – continuous feedback loops that refine policy and technical design.

  • Cited India’s FinTech, digital‑health, and open‑network pilots that demonstrated reduced uncertainty and sustained innovation velocity.

2.4 Guiding Principles for AI‑DPI Experimentation

  • Human‑centric design – placing citizen welfare at the centre.
  • Safeguards, interoperability, open standards – ensuring solutions can plug into the broader India Stack.
  • Capacity & institutional readiness – requiring policymakers, technologists, and frontline administrators to learn through sandbox trials.

2.5 Call to Action

  • Announced the launch of India’s AI‑governance framework and invited global innovators to contribute risk‑assessment methodologies.

3. Video / Report Launch (Lorraine Porciuncula, DataSphere Initiative)

3.1 Contextual Introduction

  • Lorraine introduced herself (via pre‑recorded video) and thanked the organizers. She referenced the pre‑summit dialogue (Dec 2023) on digital trust.

Key Takeaways

  1. AI and DPI are intertwined infrastructures – they must be designed together.
  2. Trust building and inclusive participation are foundational.
  3. Experimentation and iteration must become the norm for responsible AI‑DPI development.

3.3 The Role of DPI Sandboxes

  • Defined DPI sandboxes as “laboratories of trust” that provide technical testing and safeguard validation while fostering collective learning.
  • Noted sandboxes cannot solve all governance problems but can surface trade‑offs, test safeguards, and generate actionable evidence before large‑scale rollout.

3.4 Report Overview – “Sandboxes for DPI: Co‑creating the Blocks of Digital Trust”

  • Structured into three sections: What are sandboxes for DPI?, Why are they needed?, Where do they exist?
  • Introduced a taxonomic framework distinguishing regulatory, operational, and hybrid sandboxes, and presented the first formal definition of DPI sandboxes (focused on identity, payments, and data‑exchange layers).

3.5 Global Landscape

  • Mapped 16 pioneering DPI sandbox initiatives (India, Europe, Africa, etc.).
  • Highlighted common success factors: feedback loops, institutional learning, and hybrid governance models.
  • Observed a prevalence of sandbox activity in digital‑identity systems, followed by data‑exchange and payment layers.

3.6 Recommendations

  • Position sandboxes as institutional capabilities, not one‑off pilots.
  • Emphasise continuous learning, rights‑based safeguards, cyber‑resilience, and digital sovereignty.
  • Announced a 2026 toolkit and a sandbox summer school (Lisbon) to translate findings into practice.

3.7 QR‑Code Announcement

  • Displayed a QR‑code for attendees to download the full report; the moderator later reminded the audience to scan it.

4. Panel Discussion – Moderated by Sushant Kumar

The panel featured seven speakers representing governments, multilateral organisations, civil‑society, and research. The discussion was structured around a series of thematic questions.

4.1 Tanzania – Digital Foundations & AI Aspirations (Dr. Nkundwe Moses Mwasaga)

  • Historical trajectory: ICT transformation began in 2003, moving from data digitisation → digitalisation → full digital transformation.

  • Three pillars of Tanzania’s DPI:

    1. Digital ID (“Jamii Number”) – creates a digital citizen base.
    2. Interoperability bus (“Jamii Bus”) – over 800 government systems interconnected.
    3. Instant payment system – enables seamless transactions, targeting the large informal sector.

  • AI vision: personal‑governance (AI‑driven personalised services) to unlock entrepreneurial opportunities.

  • Challenges identified (rapid‑fire):

    1. Digital skills for a 62 M population.
    2. Digital security & trust – cybersecurity, privacy, consumer protection.
    3. Connectivity – expanding 4G/5G coverage.
    4. Institutional capacity across five pillars (skills, security, telecom, digital economy, entrepreneurship).

4.2 Nepal – Balancing Innovation & Regulation (Adesh Khadka)

  • Current DPI status: national ID largely enrolled; vibrant private‑sector payment ecosystem; missing a robust data‑access platform.

  • Policy backdrop: newly adopted AI policy that explicitly calls for sandboxing.

  • Two key governance questions:

    1. Capacity of regulators – Are they aware of emerging risks?
    2. Acceptance of failure – How to embed a culture that tolerates experiment‑driven failures?

4.3 UNDP Asia‑Pacific (Alexandru Oprunenco)

  • Common government query: “What is the concrete use‑case?” – many officials cannot articulate a clear problem statement.

  • Three guiding questions UNDP uses:

    1. Why – the policy intent behind DPI and AI.
    2. For whom – citizen‑centric vs. internal bureaucratic or private‑sector benefits.
    3. What changes – how sandbox outcomes reshape inter‑agency relationships and institutional learning.

  • Emphasised that sandbox design must consider organisational impact, not just technical validation.

4.4 Africa Sandboxes Forum – Hands‑On Coaching (Morine “Maureen” Amutorine)

  • Highlights the “challenge‑sandboxable?” diagnostic: many African ministries recognise sandbox value but lack a concrete, testable problem.
  • Describes one‑on‑one coaching to help governments move from idea to execution, providing “play‑cards” for sandbox planning, execution, and evaluation.
  • Reinforces the report’s definition: controlled learning environments with governance, safeguards, and time‑boxes.

4.5 Governance & Measurement (Dr. Verena Kontschieder)

  • Stresses contextual testing: AI cannot be isolated from DPI; must be evaluated alongside identity, payments, and data‑exchange layers.
  • Warns of foundational bias – if underlying DPI data are biased, AI amplifies inequities.
  • Sandboxes enable multi‑stakeholder engagement (government, innovators, civil society) to surface rights‑based concerns, cybersecurity, sustainability, and cross‑border data sovereignty.

4.6 Synthesis – Shared Themes

ThemeInsight
Use‑case clarityGovernments often lack a concrete AI‑DPI problem; sandbox design should start with a well‑defined question.
Human‑centric focusThe “for whom” question repeatedly surfaces: citizen benefit must trump bureaucratic convenience.
Iterative learningSandboxes act as labs of trust, producing evidence that informs policy, regulation, and institutional change.
Risk anticipationEarly detection of bias, privacy breaches, and systemic inequities is a core sandbox benefit.
Capacity buildingBoth regulator expertise and broader digital skills are essential to sustain sandbox outcomes.
Collaboration across sectorsEffective sandboxes blend private‑sector innovation, civil‑society watchdogs, and government oversight.

4.7 One‑Word Reflection Exercise

  • Moderator asked each panelist to share one word that encapsulated the discussion. Responses (in order of appearance) were:

    • Alexandru Oprunenco – Experiment
    • Adesh Khadka – Sustainable
    • Morine Amutorine – Trust, Confidence, Citizen‑centric (three words)
    • Dr. Nkundwe – Trust
    • Kavita Bhatia (quoted by moderator) – Trust

  • The exercise underscored trust and experimentation as the central pillars.

4.8 Final Provocative Question – “Are governments solving for themselves or for citizens?”

  • Adesh Khadka answered: governments often have dual motives – serving citizens while also satisfying internal power dynamics and private‑sector interests. True adoption requires aligning these motives.

  • Dr. Nkundwe provided a checklist for governments:

    1. What problem does DPI aim to solve, and whose problem is it?
    2. What new problems could DPI create?
    3. Who is most likely to be harmed by DPI?

  • Consensus: citizen‑centric framing must guide every sandbox experiment.

5. Closing Remarks

  • The moderator thanked the panel and reminded attendees to scan the QR‑code for the DataSphere report.
  • Noted that audience Q&A was not possible due to time constraints, encouraging participants to approach speakers after the session.
  • Announced a souvenir from the Impact Summit and officially closed the session.

See Also:

Graph View

Backlinks