25 june 2026 – ‘s-Hertogenbosch

As organisations increasingly adopt agentic AI, striking the right balance between autonomy and accountability becomes critical. While autonomous systems can significantly increase speed and efficiency, they can also challenge traceability, control, and oversight.

This session focuses on establishing a clear governance framework for decision-making, escalation, and ownership. From the outset, organisations must define which decisions remain non-delegable and always require human accountability, in some cases even at board level. At the same time, effective governance demands clear processes that enable rapid decision-making and escalation without compromising control.
Maintaining control over agentic AI requires both clarity in roles and responsibilities and the ability to act swiftly when needed. These capabilities are essential to ensure accountability, manage risk, and unlock the value of AI in a responsible and controlled way.

Rens and Dimitri will explore how organisations can prevent AI agents from “derailing” their defenses. Even agents that pass initial security reviews can introduce hidden risks at runtime through prompt injection, tool poisoning, or destructive over-alignment. This creates a false sense of security: systems appear safe but carry vulnerabilities that only emerge in operation.

Unlike traditional software, agentic AI operates probabilistically, blurs security boundaries, and may incorporate instructions from both trusted and untrusted sources. These characteristics require a new approach to security and governance.

Drawing on AI governance practices at Dutch Railways, real-world incidents, and the latest research, Rens will demonstrate how agentic AI is reshaping software development and maintenance, and which defense-in-depth strategies enable organisations to govern AI agents effectively without stifling innovation.

Speaker(s)