session Replay

AI Attack Lab: Breaching AI Agents & Chatbots

Watch this live simulation, where we walked through three real-world AI attack patterns from the hacker’s perspective. You’ll see how “normal” chatbot and agent behavior can quietly be redirected to unauthorized actions – and how to detect and block those threats in real-time using runtime guardrails and agentic policies.

What we cover...

  • A red-team walkthrough of three AI vulnerabilities affecting agents and chatbots
  • How tool poisoning can introduce hidden instructions and unauthorized behavior
  • How externally sourced content can carry indirect prompt injection “instructions” into chatbots
  • How to identify agentic threats and risky behaviors within your AI stack

  • Runtime guardrails to prevent prompt injection, data exfiltration, and unsafe actions

Presenter:

Abdiel Santos - clear

Abdiel Santos

Customer Success Manager, Varonis