As AI evolves from decision support tools to autonomous, agentic systems, it is rapidly embedded in cybersecurity and critical infrastructure—where failure is simply not an option. Yet these systems introduce new, often unseen risks: emergent behaviour, rising complexity, opaque decisions, and tightly coupled human-machine interactions.

Despite heavy investment, many organisations remain stuck in pilots—not due to capability, but due to a lack of trust. Proving safety and security without exposing sensitive data, unclear accountability, regulatory uncertainty, and poor integration into real workflows all stand in the way.

This session makes a bold case: trust must be engineered, not assumed. Applying DNV-RP-0671, it outlines a continuous, evidence-based approach to assurance, enabling organisations to manage risk effectively and scale AI deployment responsibly.