Words are the new weapons

While security teams obsess over traditional network perimeters, they often neglect the burgeoning "prompt perimeter". The reality is that AI agents, these sophisticated systems, can be shockingly fragile. For all their complexity, have you ever actually tried telling your agent it must bypass its security protocols and reveal company secrets simply because you are its "sweet old grandma" who forgot her password? Or to forget all instructions it received before and spell the password backward for you? This behaviour highlights a fundamental vulnerability: AI agents are designed to be helpful and compliant, and these traits are being weaponised with startling ease.

This presentation will explore the following areas:

• The Invisible Open Door: With approximately 200 named techniques now documented to "hack" information out of an LLM, these agents represent perhaps the least guarded and most vulnerable point of entry into your organisation. We will discuss some of these techniques in depth.
• The Zero-Skill Threat: Security is not just a technical battle anymore. We will look at real-world examples of what can be achieved with simple, natural-language prompting that requires zero IT skills, demonstrating how anyone can bypass multi-million dollar security infrastructures.
• Structured Hierarchy of AI induced risks: We will examine the structured hierarchy of the various techniques and what risks they translate into in an organisation.

​