Shone Mousseiri

As generative AI moves into production, the threat landscape is shifting from system breaches to manipulation of model behaviour through adversarial attacks such as prompt injection and jailbreaks. Traditional cyber controls, designed for deterministic systems, struggle to defend against these dynamic, context-driven risks. In this session, Shone Mousseiri, Head of Global AI Validation and Governance, and Angel Chang, Senior AI Governance Manager at Manulife, will explore why generative AI amplifies adversarial risk and why existing cyber defenses alone are insufficient. The discussion will examine how red teaming and model validation serve as critical assurance mechanisms, testing not only whether systems are secure, but whether they behave as intended under realistic and adversarial conditions. Drawing on real-world practice, the session will also explore practical approaches to operationalizing red teaming across the AI lifecycle.

• Understanding adversarial risks in generative AI environments.

• Implementing red teaming across the AI lifecycle.

• Strengthening model assurance through continuous validation.