Introduction

Turing laureate Yoshua Bengio sits down at the AI for Good Global Summit 2026 in Geneva to talk about building safety into AI development. The conversation covers the societal and technical safeguards needed to reduce risks from advanced AI — and how safety can be built more directly into the systems themselves.

Capability Brings Responsibility

Bengio argues that building more capable systems means confronting the risks that come with them. The more powerful the tool, the more seriously we must take the duty to keep it safe — and the harder that duty becomes.

Safety by Design

The central argument: safety should be built directly into development, not bolted on after the fact. Waiting until a system is deployed and then adding guardrails is too late — the architecture itself has to be safe.

Technical and Societal Safeguards

Bengio describes two layers of protection working together: the technical measures built into the models, and the societal guardrails — laws, norms, and oversight — that surround them. Neither layer alone is enough.

Global Risks, Global Cooperation

Recorded on an ITU and UN stage, the conversation is itself an act of global cooperation. AI risks do not stop at borders, and mitigating them requires the international community to work together — a theme the AI for Good Summit embodies.

The Voice Up Connection

Voice Up includes this conversation because precision means hearing the people building the safest systems alongside the people calling for the strongest restrictions. Hold Bengio's safety-by-design case next to Gates's call for required safeguards — and find your own purpose in the tension.

The Naming Gap

Safety by design is a Naming Gap question: do we name the real mechanism (safety built into the architecture) or the comforting symbol (a guardrail added later)? Systems fail when they recognize the symbol instead of the mechanism.

Make the Difference

Building safety into development is environment design. Bengio wants safety woven into the architecture the way Voice Up builds purpose into a day — not as an emergency brake, but as the environment itself.

The Purpose Games

Global cooperation on AI risk is a long game. The AI for Good Summit is one move in a negotiation that will take years. Purpose is the practice of staying at the table for as long as agreement takes.