#1 - Summits season

Key points:

The Questions:

• What will meaningful policy recommendations on advanced AI and safety look like?

• What does a global governance agenda on the safety of advanced systems look like from an Indian perspective?

• Will Bern host the 2027 Summit?

Word of this Edition: Liminality

As I start this post, France’s new PM just resigned. This reminds me of the political chaos France has experienced since June 2024, when Emmanuel Macron dissolved the National Assembly, effectively waving goodbye to any effective ruling majority.

Today is a great day to share some thoughts on AI global governance, which was recently discussed…

In New York…

Let’s travel back to Bletchley Park 2023 (UK) for a second here.

At the first AI Safety Summit in the UK, countries agreed on the need to improve States’ understanding of the capabilities of advanced AI systems, the harms and risks they posed, and where gaps existed in safety research.

This agreement led to the International AI Safety Report, the “world’s first comprehensive synthesis of current literature of the risks and capabilities of advanced AI systems”.

The mandate of this report was clear: technical report only, no policy recommendations. Yet, people have always been asking for policy recommendations. For example, it is interesting to know that “tests of scientific reasoning and programming have generated new evidence for potential risks such as AI-enabled hacking and biological attacks,” but as a State, what should you do about it?

Well…the policy vacuum is being filled from multiple directions. Maria Ressa recently presented the Red Lines initiative, which aims to establish concrete guardrails for advanced AI systems. The most recent IDAIS US-China Dialogue offers one example of what actionable recommendations on red lines might look like — defining clear boundaries that both nations could adopt to prevent AI-related risks. These initiatives represent practical attempts to translate technical insights into policy action.

The UN is also stepping in. In late September 2025, the international community in New York agreed to launch (a) a scientific panel to help bridge AI’s tech and policy gaps and (b) a global dialogue on AI governance.

The scientific panel will comprise 40 members, serving in their personal capacity for a three-year term beginning in 2026. It will play a key role in advancing global, evidence-based understanding of artificial intelligence. Its outputs should not be prescriptive, but may be policy-relevant.

So, the question is: What will meaningful policy recommendations on advanced AI and safety look like, whether from the UN panel or alternative initiatives?

The International Safety Report on advanced AI is being revamped to be more concrete and actionable. We expect a key update on recent technical developments on October 15th — more soon.

The 2026 edition should, among other new items, include forecasting scenarios and track improvements in interpretability, reliability and controllability.

And it should be presented…

…In Delhi

As was mentioned in Paris in February, the next AI Summit will take place in India, on February 19th and 20th. It should be the platform for the 2026 International Safety Report.

India is the first Global Majority country to host the Summit, following the UK (2023), South Korea (2024) and France (2025). The IndiaAI Impact Summit has the following objectives:

• Showcasing India’s AI prowess to the world;

• Harnessing AI for social good and development;

• Aligning global discourses on AI to needs of the developing world;

• Democratizing AI and bridging the AI divide;

• Positioning India as a leading voice of the Global South on AI.

In other words, the Summit aims to (a) reinforce India’s role as a global tech power, (b) highlight work done by its public infrastructures and private entities and (c) tell a story that fits its national needs without upsetting allies.

The question I therefore ask is: what does a global governance agenda on the safety of advanced systems look like from an Indian and global majority perspective?

One could also wonder how Summit hosts could strengthen continuity and institutional memory across Summits. In this paper, authors (of which I am one) argued that a Secretariat could help States adopt a hybrid model with a semi-permanent core team retained across multiple summits to ensure institutional memory and stable operations, while allowing hosts the flexibility to shape agendas and invite participants.

In any case, continuity between Summits is something that might be on the minds of diplomats…

…In Bern

Why?

Last June, Switzerland’s Federal Council announced that Switzerland would like to host an international summit on AI in Geneva in 2027.

It has mandated the Federal Department of the Environment, Transport, Energy and Communications (DETEC) and the Federal Department of Foreign Affairs (FDFA) to begin preparations, in close coordination with India.

Switzerland’s objectives in hosting the 2027 Summit seem to be (a) strengthening its engagement in international discussions on AI governance and (b) highlighting its role as a center of innovation and research.

More soon(ish).

Closing Word

Word of this Edition: Liminality

To Read:

• What Kind of Great Power Will India Be? Debating New Delhi’s Grand Strategy, Nirupama Rao; Dhruva Jaishankar; Lisa Curtis; Ashley J. Tellis, Foreign Affairs, July 30, 2025

• We must build AI for people; not to be a person, Mustafa Suleyman, August 19th, 2025

To Listen:

• Joe Hisaishi, A symphonic celebration, Music from the Studio Ghibli Films of Hayao Miyazaki, Royal Philarmonic Orchestra