Building Momentum: From Delhi to Geneva

As we wrap up the fourth international AI Summit, which took place last week in New Delhi, India, our team has been reflecting with our partners on the issues that surfaced throughout the week: issues reflecting global trends as well as the concerns and hopes from the global AI community.

The India AI Summit, organised under the IndiaAI Mission by the Ministry of Electronics and Information Technology, marked a significant moment as the first summit in the series to be hosted by a Global South nation. This milestone in the global responsible AI conversation reaffirmed the necessity of bringing together a broad set of stakeholders spanning the entire AI value chain — from developers to end users — and ensuring meaningful representation from AI hubs across the world.

At the Summit, PAI held an event as part of the official proceedings on Preparing to Monitor the Impacts of Agents. We were honored to have keynotes from Minister Josephine Teo of Singapore and Natasha Crampton from Microsoft. Our expert panel was moderated by Madhu Srikumar from PAI and included Fred Werner of the ITU, Owen Larter from Google DeepMind, Vukosi Marivate from the PAI Board and the University of Pretoria and Stephanie Ifayemi of PAI. Chris Meserole of the Frontier Model Forum provided closing remarks. See the video here. As part of this event, we released two reports on AI Assurance.

Throughout these convenings and with the launch of the Delhi Declaration, four key takeaways stood out:

1. Standardized Monitoring for Accountability and Transparency

Rapid AI adoption needs to be coupled with real-time standardized monitoring to ensure progress translates into shared prosperity. The goal is to move beyond “black box” systems and toward greater transparency, empirical data and accountable tools that allow governments, researchers, and communities to test, verify, and improve outcomes dynamically across the AI value chain.

To that end, the Delhi Declaration’s emphasis on transparency and the use of real-world usage data to understand AI’s impacts reflects the recommendations in our research to document the impact of foundation models, and where our work has been heading for years.

2. Cultivating Diverse Ownership Across the AI Stack

The Delhi Declaration emphasises strengthening multilingual models through intentional diversification to ensure the AI ecosystem genuinely reflects the world’s full range of cultural, linguistic, and local contexts. This means moving beyond surface-level inclusion to embedding representative data at the foundation of AI development, and ensuring that communities and countries hold meaningful ownership at every layer of the AI value chain.

This requires candid discussion about autonomy across the stack: who owns data, who controls local deployment, and which countries have the infrastructure capacity to participate as genuine partners rather than passive recipients. Solutions designed without local input risk deepening existing divides. Greater investment in shared or sovereign infrastructure and clearer frameworks for data ownership are not peripheral concerns — they are foundational to whether AI development is truly diverse.

Moving forward, deeper discussions are needed on how to implement this objective — leveraging open-source solutions, understanding accountability chains and their related assurance regimes, and building meaningful partnerships. In doing so, we can shape an AI landscape that is both technically advanced and truly representative.

3. Intentional & International Collaboration

In a shifting global order where traditional multilateralism and governance mechanisms face unprecedented challenges, collaborative partnership remains the only viable path to establishing consistent safety standards. This collective approach is essential to ensure AI development stays secure, human-centric, and reflective of the world’s full diversity.

Navigating this landscape demands more than high-level statements — it requires active, sustained engagement across all global AI hubs and the entire AI value chain, bringing together developers, deployers, policymakers, civil society, and directly affected communities. This is why PAI is deepening its international presence and engagement, while maintaining a nimble organisational footprint by targeting highly influential regional and sectoral actors and prioritising the depth and quality of its partnerships.

4. The Road to Geneva 2027

The momentum from New Delhi needs to carry forward to the next Global AI Summit in Geneva in 2027 and beyond.
A year ago, at the Paris AI Action Summit, we identified a clear policy gap when it came to defining the core functions of robust AI assurance. Following the Summit, we launched a yearlong, multistakeholder effort that resulted with the release of a landmark report, Strengthening the AI Assurance Ecosystem, in Delhi.

As we look to Geneva in 2027, we are excited to activate the insights in this report with our partners this year and to plan proactively for the next Summit. To that end, we are making the following two additional commitments:

• Summit Synchronization Conversations: Starting in Spring 2026, PAI will host regular zoom calls to share knowledge and align our collective expertise with the objectives of the 2027 Summit. Our goal with this Community of Practice is to provide our Swiss hosts with timely, evidence-based, and actionable recommendations that move the global dialogue beyond high-level principles toward meaningful implementation.
• Commitment Tracking: PAI will track progress toward past and future commitments made during the AI summit. We must be able to clearly state what was agreed upon in London, Seoul, Paris, and Delhi—and exactly where we stand on implementation. We will also use the 2026 ITU AI For Good Summit as well as the upcoming UN Global Dialogue as touch points with the broader community to reflect on past achievements and future efforts. This commitment builds on our existing work to drive coherence and alignment across governance efforts.

We look forward to working with the broader AI community and our partners — new and existing — to advance global commitments and strengthen accountability towards the promises made.