John O'Donnell

As organizations deploy AI agents and models to interact directly with external users, risk exposure increases across security, compliance, and trust. These systems often operate without human oversight, raising challenges around bias, prompt injection, data leakage, and unpredictable behaviour. This session focuses on governing public-facing agentic AI across the full lifecycle, with emphasis on pre-deployment testing, adversarial evaluation, and continuous monitoring post-deployment. Panellists also discuss challenges and the tension between scaling ROI and ensuring systems are secure, reliable, and accountable.

• Identifying and mitigating risks in public-facing, autonomous AI systems.
• Strengthening pre-deployment testing and adversarial evaluation practices.
• Balancing ROI with security, compliance, and lifecycle governance.