Skip to main content
back to List back
on this page

Offensive & Defensive AI: Exploiting and Securing LLMs, RAG, and Agentic Systems

live online
price $1,400
Adeel
Intermediate / Advanced
2 Days
AI Focus
Sun 6 Dec - Mon 7 Dec
price $1,400
Register


Online, Available in English and Arabic

Overview

Large Language Models (LLMs), Retrieval-Augmented Generation (RAG), and agentic systems are rapidly being deployed across enterprises, often without adequate security controls. 

This hands-on training focuses on the emerging AI attack surface, teaching participants how to exploit and secure AI systems. Participants will build AI workflows, perform real-world attacks such as prompt injection and data exfiltration, and implement practical guardrails to mitigate these risks.

This 2-day, hands-on technical training focuses on both offensive and defensive aspects of AI systems.

Day 1 focuses on building LLM-based applications, RAG pipelines, and agentic systems to understand how they work in real environments.

Day 2 focuses on attacking these systems using real-world techniques such as prompt injection, data leakage, and tool abuse, followed by implementing security controls and guardrails.

The course is lab-driven and uses isolated environments with controlled datasets.

Topics:

  • Building AI Systems (Understanding the Attack Surface)
  • LLM architecture (practical view for security) 
  • Prompt engineering and failure modes 
  • RAG pipelines and data flow 
  • Agentic AI systems (tools, memory, execution) 
  • Attack surface mapping for AI systems 
  • Lab 1: Build a basic LLM application
  • Lab 2: Build a RAG pipeline with document retrieval
  • Lab 3: Create a simple AI agent with tool access
  • Prompt injection (direct and indirect) 
  • Data exfiltration via LLMs 
  • Retrieval poisoning (RAG attacks) 
  • Tool abuse and privilege escalation in agents 
  • Multi-agent risk scenarios 
  • Lab 4: Exploit LLM using prompt injection
  • Lab 5: Execute data leakage attack via RAG
  • Lab 6: Abuse agent tools for unintended actions
  • Hardening & Defense
  • Instruction vs data separation 
  • Input/output validation 
  • Access control for tools and retrieval 
  • Safe agent design patterns 
  • Monitoring and detection for AI misuse 
  • Lab 7: Apply guardrails and re-test attacks

Target audience and prerequisites

Security engineers, AI engineers, developers, SOC analysts. Basic Python and API knowledge recommended.

By the end of this course, the participant will be able to: 

  • Build LLM-based applications and RAG pipelines
  • Develop agentic AI systems with tool integration
  • Identify and exploit vulnerabilities in AI systems
  • Perform attacks such as prompt injection and data exfiltration
  • Implement security guardrails to protect AI systems.