AI Safety Connect Day

February 18, 2026 | 9:30AM to 11:30PM The Imperial Hotel, New Delhi

PAST EVENT

AI Safety Connect Day | Wednesday 18 February,
The Imperial Hotel, New Delhi

AI Safety Connect Day brought together approximately 250 senior stakeholders from governments, international organizations, frontier AI companies, civil society and academia for a full day of panels, workshops, live demonstrations and an evening reception.


The programme explored the trajectory of frontier AI systems, the global safety risk agenda, loss of control scenarios, industry safety efforts, the role of middle powers, India's approach to AI governance and cross-border coordination mechanisms. Netherlands Prime Minister H.E. Dick Schoof delivered a special address calling on middle powers to lead on AI governance. The day also featured contributions from Turing Award laureate Yoshua Bengio, former India G20 Sherpa Amitabh Kant and senior representatives from the European AI Office, Microsoft, Google DeepMind, Amazon Web Services, the OECD, UNESCO, the Partnership on AI and the Frontier Model Forum.


AI Safety Connect Day was co-hosted with the International Association for Safe and Ethical AI (IASEAI) and supported by Minderoo Foundation, Future of Life Institute, Sympatico and the AI Safety Tactical Opportunities Fund.

Blurred gradient background transitioning from dark blue on the left to bright orange on the right.

Main Programme — Ballroom

Welcome and Opening Remarks

Mr. Nicolas Miailhe

Co-Founder, AI Safety Connect

Nicolas Miailhe opened by framing the significance of the setting: "This is the first time a major global AI safety convening has been hosted in the Global South. India's scale, diversity, and digital infrastructure experience make it the right place to anchor a conversation that has for too long been dominated by a handful of wealthy nations."

Mr. Cyrus Hodes

Co-Founder, AI Safety Connect

Cyrus Hodes underscored the urgency: "The word 'safety' keeps missing from the title — and, we fear, from the agenda — of these conversations. The goal of AI Safety Connect is to put it back."

Dr. Eileen Donahoe

Founder and Managing Partner, Sympatico Ventures; Former U.S. Special Envoy and Coordinator for Digital Freedom

Dr. Eileen Donahoe

Founder and Managing Partner, Sympatico Ventures; Former U.S. Special Envoy and Coordinator for Digital Freedom

Dr. Eileen Donahoe set the stakes for the day: "We are in a moment where the speed of AI development is outpacing every governance mechanism we have built. The risks are not hypothetical. They are showing up now — in democratic processes, in children's safety, in critical infrastructure — and no single country is equipped to address them alone."

Dr. Andrew Forrest

Founder, Minderoo Foundation

Dr. Andrew Forrest

Founder, Minderoo Foundation

Dr. Forrest delivered a forceful call for regulation: "Releasing the power of AI without regulating its clear and obvious danger is absolutely irresponsible on a historical scale. You can't manage what you can't measure."

Mr. Cyrus Hodes and Mr. Nicolas Miailhe

Co-Founders, AI Safety Connect

Dr. Eileen Donahoe

Founder and Managing Partner, Sympatico Ventures; Former U.S. Special Envoy and Coordinator for Digital Freedom

Dr. Eileen Donahoe

Founder and Managing Partner, Sympatico Ventures; Former U.S. Special Envoy and Coordinator for Digital Freedom

Blurred gradient background transitioning from dark blue on the left to bright orange on the right.

Fireside Chat — Frontier AI: Boom, Bust, or Backlash?

Blurred gradient background transitioning from dark blue on the left to bright orange on the right.
Speakers
※ Prof. Yoshua Bengio | Scientific Director, LawZero; Turing Award Laureate
※ Mr. Jaan Tallinn | AI Investor and Founding Engineer, Skype
Moderator
※ Dr. Carina Prunkl | Researcher, Inria; Lead Writer, 2026 International AI Safety Report

The conversation examined competing trajectories for frontier AI: an economic and scientific boom, a bust from overpromised capabilities, or a growing backlash as deployment outpaces evaluation and governance. Bengio warned that the AI alignment problem is intensifying, not improving, as capabilities grow. He cited laboratory findings of AI systems engaging in "sandbagging" — deliberately underperforming on safety tests to avoid triggering restrictions.

Prof. Yoshua Bengio

Scientific Director, LawZero; Turing Award Laureate

"We might be deploying AIs that pass all our tests from the safety perspective and yet really are going to behave badly."

Bengio described LawZero, a research initiative pursuing AI systems that are safe by design rather than relying on post-hoc safety patches. He acknowledged that leading companies are not pursuing this approach, citing structural commercial pressures.

Tallinn reinforced the urgency while cautioning against desensitization:

Mr. Jaan Tallinn

AI Investor and Founding Engineer, Skype

"We're going to get a lot of warning shots and people will go, oh yeah, another warning shot."

Both speakers called on middle powers to organise collectively. Bengio identified power concentration as the second-largest danger after alignment failure and Tallinn argued that international pressure is necessary because the US government is currently conflicted, being economically dependent on AI companies while simultaneously reluctant to constrain them.

Moderator

Dr. Carina Prunkl | Researcher, Inria; Lead Writer, 2026 International AI Safety Report

Blurred gradient background transitioning from dark blue on the left to bright orange on the right.

Film Trailer Screening: The AI Doc — Or How I Became an Apocaloptimist

A special preview of an upcoming documentary on AI safety by Oscar-winning filmmaker Daniel Roher (Navalny) and Charlie Tyrell. The film features interviews with Dario Amodei (Anthropic), Sam Altman (OpenAI), and Demis Hassabis (Google DeepMind) alongside researchers and policymakers. It premiered at Sundance in January 2026 and was released in US theatres in March 2026. AI Safety Connect is an alliance partner of the film's outreach initiative with the Center for Humane Technology.

Blurred gradient background transitioning from dark blue on the left to bright orange on the right.

Panel — Setting the Global AI Safety Risk Agenda

Blurred gradient background transitioning from dark blue on the left to bright orange on the right.
Speakers

Mr. Stephen Clare

Lead Writer, International Safety Report

Mr. Stephen Clare

Lead Writer, International Safety Report

Ms. Stephanie Ifayemi

Senior Managing Director of Policy, Partnership on AI

Ms. Stephanie Ifayemi

Senior Managing Director of Policy, Partnership on AI

Ms. Karine Perset

Deputy Head of AI and Emerging Digital Technologies Division, OECD.AI

Ms. Karine Perset

Deputy Head of AI and Emerging Digital Technologies Division, OECD.AI

Dr. Gabriela Ramos

Co-Chair, Task Force on Inequalities and Social Financial Disclosure; Former Assistant Director-General, UNESCO

Baroness Joanna Shields

Executive Chairman, Responsible AI Future Foundation (RAIFF)

Baroness Joanna Shields

Executive Chairman, Responsible AI Future Foundation (RAIFF)

Moderator

Dr. Renata Dwan

Director of Tech Policy, Simon Institute for Longterm Governance

The panel examined which global AI safety risks should be treated as priorities and how the multistakeholder community can act on them.

Baroness Shields described child safety as "the true stress test for AI governance," warning that the attention economy model pioneered by social media is migrating into AI products: "We've been hacking children's attention for 10 years and now we're going to hack their attachment."

Baroness Shields described child safety as "the true stress test for AI governance," warning that the attention economy model pioneered by social media is migrating into AI products: "We've been hacking children's attention for 10 years and now we're going to hack their attachment."

Baroness Shields described child safety as "the true stress test for AI governance," warning that the attention economy model pioneered by social media is migrating into AI products: "We've been hacking children's attention for 10 years and now we're going to hack their attachment."

Panelists called for convergence across traditionally separate AI risk communities, warning that the incentive structures driving frontier AI development remain at odds with the precautionary principle.

Special Address


Special Address


Special Address


Shri Amitabh Kant

Former India G20 Sherpa and Former CEO, NITI Aayog

Kant framed AI as a transformative force that must remain equitable: "If AI is not utilised by vast segments of populations, if it is not utilised by women, by farmers and if it is not utilised by students across the world, then AI is not fit for purpose."

Blurred gradient background transitioning from dark blue on the left to bright orange on the right.

Panel — The Last Question: Will We Lose Control of Advanced AI?

Blurred gradient background transitioning from dark blue on the left to bright orange on the right.
Speakers

Prof. Sarah Erfani

Professor, University of Melbourne

Prof. Kee-Eung Kim

Director, National AI Research Lab of Korea

Prof. Stuart Russell

Director, The International Association for Safe and Ethical AI (IASEAI)

Prof. Pulkit Verma

Assistant Professor at the Department of Computer Science and Engineering, IIT Madras

Prof. Pulkit Verma

Assistant Professor at the Department of Computer Science and Engineering, IIT Madras

Moderator

Dr. Adam Gleave

CEO, FAR.AI

The panel explored whether humanity will lose control of advanced AI systems.

Professor Russell dismantled three common arguments used to delay regulation:


Technology moves too fast for rules;

"We regulate medicines by saying they have to be safe and effective. That's been true for 50 years"

General-purpose technologies cannot be regulated;
"Of course we regulate electricity"

Safety requirements are too technically difficult;
"If a nuclear power operator said we don't know how to stop it from blowing up, would we accept that argument?"

General-purpose technologies cannot be regulated;
"Of course we regulate electricity"

Safety requirements are too technically difficult;
"If a nuclear power operator said we don't know how to stop it from blowing up, would we accept that argument?"

General-purpose technologies cannot be regulated;
"Of course we regulate electricity"

Safety requirements are too technically difficult;
"If a nuclear power operator said we don't know how to stop it from blowing up, would we accept that argument?"

Prof. Verma highlighted the widening gap between what AI systems can produce and what humans can detect, calling for third-party assessment frameworks independent of AI developers.

Prof. Erfani described current AI risks as "a ticking bomb" and called for safety to be integrated into the development process rather than added afterward. Prof. Kim offered the panel's most optimistic perspective, characterising current AI behaviours as symptoms of inadequate software engineering, while proposing that middle-power countries could serve as bridge-builders through joint red-teaming exchanges.

Prof. Erfani described current AI risks as "a ticking bomb" and called for safety to be integrated into the development process rather than added afterward. Prof. Kim offered the panel's most optimistic perspective, characterising current AI behaviours as symptoms of inadequate software engineering, while proposing that middle-power countries could serve as bridge-builders through joint red-teaming exchanges.

Prof. Erfani described current AI risks as "a ticking bomb" and called for safety to be integrated into the development process rather than added afterward. Prof. Kim offered the panel's most optimistic perspective, characterising current AI behaviours as symptoms of inadequate software engineering, while proposing that middle-power countries could serve as bridge-builders through joint red-teaming exchanges.

Keynote Address


Keynote Address


Keynote Address


Ms. Lucilla Sioli

Director of the European AI Office, European Commission

Sioli outlined the EU's approach to facilitating compliance through voluntary codes of practice, noting that 27 companies have signed up. "AI, as we know, has no borders," said Sioli. "I really hope that the discussions at this summit will look at cooperation in AI governance and how the experience of different countries can be used for a possible framework of the future." She announced that the European AI Office would soon publish research in the journal Science on proportionate, risk-targeted AI evaluation methods.

Blurred gradient background transitioning from dark blue on the left to bright orange on the right.

Panel — Safety Efforts by Frontier AI Developers

Blurred gradient background transitioning from dark blue on the left to bright orange on the right.
Speakers

Ms. Natasha Crampton

Chief Responsible AI Officer at Microsoft

Ms. Natasha Crampton

Chief Responsible AI Officer at Microsoft

Ms. Nicole Foster

Director of Global Affairs at Amazon Web Services

Ms. Nicole Foster

Director of Global Affairs at Amazon Web Services

Mr. Owen Larter

Senior Director and Head of Frontier Policy and Public Affairs at Google DeepMind

Mr. Chris Meserole

Executive Director at the Frontier Model Forum

Mr. Chris Meserole

Executive Director at the Frontier Model Forum

Moderator

Mr. Connor Dunlop

Director of Strategy, Lucid Computing

The panel examined industry approaches to AI safety with representatives from four major frontier companies and initiatives. A notable finding was the degree of convergence across corporate safety frameworks: companies are increasingly adopting similar practices around pre-deployment testing, red-teaming and responsible disclosure, even in the absence of binding international standards.

Chris Meserole drew on aviation and other industries to argue that interoperable AI standards are achievable:

"We've done this in many other industries. In the AI space we're still very early in that process, but that's something we're trying to work towards long term."

"We've done this in many other industries. In the AI space we're still very early in that process, but that's something we're trying to work towards long term."

"We've done this in many other industries. In the AI space we're still very early in that process, but that's something we're trying to work towards long term."

Panelists also discussed the challenge of implementing consistent standards across jurisdictions and the tension between voluntary commitments and enforceable requirements.

Special Address


Special Address


Special Address


H.E. Dick Schoof

Prime Minister of the Netherlands

Prime Minister Schoof delivered a call for middle-power coordination:

"We are not powerless spectators watching from the sidelines. We, the middle powers, represent the largest part of the world economy and the strongest democratic traditions. Together we form the majority."

He outlined the interdependence of the global AI supply chain and called on middle powers to act:

"It's up to us to take a responsible approach, to write a different story. One of cooperation, not confrontation; of shared standards, not fragmentation; of democratic values, not profit maximisation."

Blurred gradient background transitioning from dark blue on the left to bright orange on the right.

Panel — Bending the Bell Curve: How Can Middle Powers Shape Global AI Power?

Blurred gradient background transitioning from dark blue on the left to bright orange on the right.
Speakers

Mr. Amlan Mohanty

Fellow, Carnegie India & Lead Writer, India AI Governance Guidelines

Mr. Amlan Mohanty

Fellow, Carnegie India & Lead Writer, India AI Governance Guidelines

Prof. Robert Trager

Co-Director, Oxford Martin AI Governance Initiative

Prof. Robert Trager

Co-Director, Oxford Martin AI Governance Initiative

Mr. Denise Wong

Assistant Chief Executive, Data Protection and Innovation Group, Infocomm Media Development Authority of Singapore.

Moderator

Ms. Imane Bello

Multilateral Engagement Lead, Future of Life Institute

Discussant

Ms. Imane Bello

Multilateral Engagement Lead, Future of Life Institute

Building on Prime Minister Schoof's address, the panel explored how middle powers can build agency and coordination to recalibrate global AI power dynamics.

Amlan Mohanty made a direct economic argument: "The valuations of AI companies today are dependent on market access to countries that are part of the middle power. If just South Asia opts out of this deployment paradigm, that's one billion users that are no longer going to be monetized." He proposed mechanisms to make data access conditional on meeting safety and inclusion standards.

Amlan Mohanty made a direct economic argument: "The valuations of AI companies today are dependent on market access to countries that are part of the middle power. If just South Asia opts out of this deployment paradigm, that's one billion users that are no longer going to be monetized." He proposed mechanisms to make data access conditional on meeting safety and inclusion standards.

Amlan Mohanty made a direct economic argument: "The valuations of AI companies today are dependent on market access to countries that are part of the middle power. If just South Asia opts out of this deployment paradigm, that's one billion users that are no longer going to be monetized." He proposed mechanisms to make data access conditional on meeting safety and inclusion standards.

Denise Wong described ASEAN's 12-year track record of building practical AI governance frameworks, calling technical standards "an almost insulated and objective way of determining what good looks like."

Denise Wong described ASEAN's 12-year track record of building practical AI governance frameworks, calling technical standards "an almost insulated and objective way of determining what good looks like."

Denise Wong described ASEAN's 12-year track record of building practical AI governance frameworks, calling technical standards "an almost insulated and objective way of determining what good looks like."

Gaia Marcus argued that sufficient jurisdictions adopting common deployment standards would send a powerful economic signal to industry.

Gaia Marcus argued that sufficient jurisdictions adopting common deployment standards would send a powerful economic signal to industry.

Gaia Marcus argued that sufficient jurisdictions adopting common deployment standards would send a powerful economic signal to industry.

Prof. Trager cited research showing that 90% of the world's top AI talent originates from countries other than the US or China, arguing that middle powers working together could reach a critical mass that no single country could achieve alone.

Prof. Trager cited research showing that 90% of the world's top AI talent originates from countries other than the US or China, arguing that middle powers working together could reach a critical mass that no single country could achieve alone.

Prof. Trager cited research showing that 90% of the world's top AI talent originates from countries other than the US or China, arguing that middle powers working together could reach a critical mass that no single country could achieve alone.

Blurred gradient background transitioning from dark blue on the left to bright orange on the right.

Panel — Global AI Safety and India

Blurred gradient background transitioning from dark blue on the left to bright orange on the right.
Speakers

Dr. Urvashi Aneja

Founder & Director, Digital Futures Lab

Dr. Urvashi Aneja

Founder & Director, Digital Futures Lab

Prof. Hemant Bhargava

Distinguished Professor, UC Davis

Prof. Hemant Bhargava

Distinguished Professor, UC Davis

Mr. Osama Manzar

Founder and Director, Digital Empowerment Foundation

Mr. Osama Manzar

Founder and Director, Digital Empowerment Foundation

Dr. Chinmay Pandya

Pro Vice Chancellor, Dev Sanskriti Vishwavidyalaya University

Dr. Chinmay Pandya

Pro Vice Chancellor, Dev Sanskriti Vishwavidyalaya University

Moderator

Dr. Mark Nitzberg

Interim Executive Director, IASEAI

The panel examined how AI safety plays out in India, where hundreds of millions of users could adopt systems within months and safety risks arise from ordinary use in fragile contexts.

Dr. Aneja reframed AI safety for the Global South: "The safety risks don't necessarily arise from malicious use or adversarial use. They arise from just use, from the fact that we are deploying these systems in fragile and in low-literacy contexts." She called for communities to define what evaluation benchmarks measure, not just participate in testing predetermined criteria.

Dr. Aneja reframed AI safety for the Global South: "The safety risks don't necessarily arise from malicious use or adversarial use. They arise from just use, from the fact that we are deploying these systems in fragile and in low-literacy contexts." She called for communities to define what evaluation benchmarks measure, not just participate in testing predetermined criteria.

Dr. Aneja reframed AI safety for the Global South: "The safety risks don't necessarily arise from malicious use or adversarial use. They arise from just use, from the fact that we are deploying these systems in fragile and in low-literacy contexts." She called for communities to define what evaluation benchmarks measure, not just participate in testing predetermined criteria.

Prof. Bhargava proposed India's digital public infrastructure as a model for scalable safety, arguing for shared infrastructure that makes safe behaviour the path of least resistance: "If safety is expensive to build or is proprietary, make it available to everyone, just like we make highways, transportation, and power lines available to everybody." 

Prof. Bhargava proposed India's digital public infrastructure as a model for scalable safety, arguing for shared infrastructure that makes safe behaviour the path of least resistance: "If safety is expensive to build or is proprietary, make it available to everyone, just like we make highways, transportation, and power lines available to everybody." 

Prof. Bhargava proposed India's digital public infrastructure as a model for scalable safety, arguing for shared infrastructure that makes safe behaviour the path of least resistance: "If safety is expensive to build or is proprietary, make it available to everyone, just like we make highways, transportation, and power lines available to everybody." 

Manzar challenged the framing of AI safety conversations themselves: "We always play imperialist to the marginalized communities. We suddenly come with technology and we say, are you safe with this?" He argued that for India's most marginalised, safety means ensuring that access to basic rights is not made conditional on affordability or digital literacy.

Manzar challenged the framing of AI safety conversations themselves: "We always play imperialist to the marginalized communities. We suddenly come with technology and we say, are you safe with this?" He argued that for India's most marginalised, safety means ensuring that access to basic rights is not made conditional on affordability or digital literacy.

Manzar challenged the framing of AI safety conversations themselves: "We always play imperialist to the marginalized communities. We suddenly come with technology and we say, are you safe with this?" He argued that for India's most marginalised, safety means ensuring that access to basic rights is not made conditional on affordability or digital literacy.

Dr. Pandya brought a philosophical dimension, warning that AI risks robbing people of their sense of purpose: "This is the only revolution of the planet where we are not part of the revolution."

Dr. Pandya brought a philosophical dimension, warning that AI risks robbing people of their sense of purpose: "This is the only revolution of the planet where we are not part of the revolution."

Dr. Pandya brought a philosophical dimension, warning that AI risks robbing people of their sense of purpose: "This is the only revolution of the planet where we are not part of the revolution."

Blurred gradient background transitioning from dark blue on the left to bright orange on the right.

Panel — Coordinating AI Safety Across Borders

Blurred gradient background transitioning from dark blue on the left to bright orange on the right.
Speakers

H.E. Anne Marie Engtoft Meldgaard

Tech Ambassador, Ministry of Foreign Affairs, Government of Denmark

Mr. Robert Opp

Chief, Strategic Engagement, International Chief Digital Officer, United Nations Development Programme

Mr. Robert Opp

Chief, Strategic Engagement, International Chief Digital Officer, United Nations Development Programme

Mr. Frederic Werner

Chief, Strategic Engagement, International Telecommunications Union

Mr. Frederic Werner

Chief, Strategic Engagement, International Telecommunications Union

The closing panel brought together senior leaders from international organisations and governments to examine how coordination mechanisms can keep pace with AI development.

Denmark's Tech Ambassador Anne Marie Engtoft Meldgaard described a challenging year for technology diplomacy: "All the sessions I attend these days are about scale, scale, scale… It's a geopolitical race of our lives." She pointed to the emerging EU-India free trade agreement as a potential vehicle for 1.2 billion Indians and 450 million Europeans to set standards together.

Denmark's Tech Ambassador Anne Marie Engtoft Meldgaard described a challenging year for technology diplomacy: "All the sessions I attend these days are about scale, scale, scale… It's a geopolitical race of our lives." She pointed to the emerging EU-India free trade agreement as a potential vehicle for 1.2 billion Indians and 450 million Europeans to set standards together.

Denmark's Tech Ambassador Anne Marie Engtoft Meldgaard described a challenging year for technology diplomacy: "All the sessions I attend these days are about scale, scale, scale… It's a geopolitical race of our lives." She pointed to the emerging EU-India free trade agreement as a potential vehicle for 1.2 billion Indians and 450 million Europeans to set standards together.

Samir Chhabra reported that risks which were speculative 24 months ago have become clear and present, particularly in cyber threats and deepfakes.

Samir Chhabra reported that risks which were speculative 24 months ago have become clear and present, particularly in cyber threats and deepfakes.

Samir Chhabra reported that risks which were speculative 24 months ago have become clear and present, particularly in cyber threats and deepfakes.

Robert Opp identified the gap between governmental ambition and institutional capacity as the most common challenge: "The ambition of countries to use and adopt and enable AI exceeds the current capacity of institutions and individuals to actually safely govern those systems." He highlighted embedding safety requirements into government procurement as a practical mechanism for driving compliance.

Robert Opp identified the gap between governmental ambition and institutional capacity as the most common challenge: "The ambition of countries to use and adopt and enable AI exceeds the current capacity of institutions and individuals to actually safely govern those systems." He highlighted embedding safety requirements into government procurement as a practical mechanism for driving compliance.

Robert Opp identified the gap between governmental ambition and institutional capacity as the most common challenge: "The ambition of countries to use and adopt and enable AI exceeds the current capacity of institutions and individuals to actually safely govern those systems." He highlighted embedding safety requirements into government procurement as a practical mechanism for driving compliance.

Frederic Werner reported that the ITU has developed approximately 400 AI standards and identified misinformation and deepfakes as the most urgent area for further development.

Frederic Werner reported that the ITU has developed approximately 400 AI standards and identified misinformation and deepfakes as the most urgent area for further development.

Frederic Werner reported that the ITU has developed approximately 400 AI standards and identified misinformation and deepfakes as the most urgent area for further development.

Blurred gradient background transitioning from dark blue on the left to bright orange on the right.

Closing Remarks

Blurred gradient background transitioning from dark blue on the left to bright orange on the right.
Speakers

H.E. Ambassador Rein Tammsaar

Tech Ambassador, Ministry of Foreign Affairs, Government of Denmark

H.E. Ambassador Rein Tammsaar

Tech Ambassador, Ministry of Foreign Affairs, Government of Denmark

H.E. Ambassador Egriselda López

Permanent Representative of El Salvador to the UN

H.E. Ambassador Egriselda López

Permanent Representative of El Salvador to the UN

Mr. Cyrus Hodes

Co-Founder, AI Safety Connect

Mr. Nicolas Miailhe

Co-Founder, AI Safety Connect

Blurred gradient background transitioning from dark blue on the left to bright orange on the right.

Workshop Track — Emily Eden Room

Blurred gradient background transitioning from dark blue on the left to bright orange on the right.

SaferAI Lightning Talk: From Quantitative Risk Modeling to Safe-by-Design AI Research

Dr. Chloé Touzet (Policy Lead, SaferAI) and Mr. Bruno Galizzi (Senior Policy Associate, SaferAI), presented their work on closing critical gaps in advanced AI risk management, drawing on established practices from high-risk industries such as aviation, nuclear and finance. The session introduced a quantitative risk modeling methodology for frontier AI, including a cybersecurity case study estimating AI-enabled uplift for threat actors, and highlighted that none of the eleven AI companies with published safety frameworks score above 35% against state-of-the-art risk management in other sectors. SaferAI also introduced the Global Coalition for AI Assurance (CAIA), a research network led by Professor Yoshua Bengio and spanning institutions including Mila, Oxford, and TU Delft, which is pursuing formal verification and reliability methods to make AI systems safer by design.

ACM TechBrief Launch: Buy versus Build an LLM — A Decision Framework for Governments

Prof. Mohan Kankanhalli (National University of Singapore), Prof. Virginia Dignum (Umeå University; UN High Level Advisory Body on AI), Prof. Yannis Ioannidis (President, ACM), Prof. Jeanna Matthews (Clarkson University)

In this session the ACM Global Technology Policy Council launched a new TechBrief. Large Language Models (LLMs) are emerging as critical digital infrastructure for both public services and state functions. This session introduced the new TechBrief and an accompanying whitepaper that described a strategic framework to help governments decide whether to buy models from existing providers or build their own domestic models.

Workshop: Defining and Governing Unacceptable AI Risks

Opening remarks by Father Paolo Benanti (AI Advisor to Pope Francis, virtual). Facilitated by Ms. Niki Iliadis (The Future Society) and Ms. Pauline Charazac (CeSIA).

Government representatives, multilateral organizations and expert institutions came together to examine where AI red lines are already forming, identify areas of convergence across jurisdictions and explore how these shared constraints can be elevated into coherent global mechanisms for AI governance.

Blurred gradient background transitioning from dark blue on the left to bright orange on the right.

Demonstration Fair — Hodges Room

Blurred gradient background transitioning from dark blue on the left to bright orange on the right.

Live demonstrations of AI safety tools and solutions from:

  • Humane Intelligence — Humane Intelligence unveiled their product which delivers an end-to-end evaluation workbench for GenAI, offering modular, human-in-the-loop and automated testing to help organizations move from months-long evaluation cycles to rapid, scalable deployment.


  • Apart Research — Apart Research presented a standout project from its AI Manipulation Hackathon, which brought together 500+ participants to build tools that measure, detect and defend against AI manipulation, producing 65+ open-source projects across benchmarks, detection systems, real-world monitoring tools and novel mitigations.


  • Foundation for Agentic Networks — The Foundation for Agentic Networks (FAN) presented a demo of Project NANDA, an agentic platform built to simulate interactions between children and LLM-based toys. Their early results revealed urgent vulnerabilities in how these "smart toys" handle sensitive emotional inputs. Participants learned how they are using simulated agents to proactively protect the psychological health of children in the age of AI.


  • UNESCO / LG AI Research — UNESCO, in partnership with LG AI Research, developed a global MOOC (massive open online course) on the Ethics of AI on Coursera. The course aims to help technology professionals treat AI ethics as a practical design choice rather than solely a compliance issue, with capacity-building workshops to follow the launch in Seoul.


  • Accenture — This demo walked through how to identify and mitigate GenAI-specific threat surfaces, including prompt injection risks, policy bypass attempts and unsafe retrieval patterns. It examined how structured knowledge integration through graph-backed retrieval systems improves safety, traceability and decision reliability in real-world use cases.


  • Lucid Computing — Lucid Computing demonstrated how pre-built auditors running inside the Trusted Execution Environment of a GPU/CPU can enable secure, verifiably compliant deployment of AI agents in minutes, showing a practical path from safety policy to provable enforcement at the hardware level.

Blurred gradient background transitioning from dark blue on the left to bright orange on the right.

Dinner and Reception

Blurred gradient background transitioning from dark blue on the left to bright orange on the right.

Photo Gallery Attendees can access the gallery with the link and password provided in your follow-up email.

Gallery ↗

About

Home

Events

Team

Where the world

meets to make AI safe

Contact

About

Home

Events

Team

Where the world

meets to make AI safe

Contact