My Talks

AI Safety Keynote · Canada

From Safe Models to Safe Deployments

A keynote on moving the AI safety conversation away from safe models and toward safe deployments — designing an assurance loop that lets model deployment be executed error-free under real conditions.

• Assurance that holds outside controlled environments, rather than only inside sandboxes.
• Containment measures that remain effective in live, real-world deployment.
• A shared reporting directory and repository where deployment errors surface — and stay legible and well-informed across the whole ecosystem.