Training > Cybersecurity > Offensive AI Exploits and Security (LFWS320)

INSTRUCTOR-LED COURSE

LLM applications introduce a new class of vulnerabilities that existing playbooks weren’t designed to address. This hands-on course gives you the specialized offensive knowledge to find, exploit, and remediate all 10 attack classes in the OWASP® Top 10 for LLM applications, from RAG prompt injections to multi-agent pipeline poisoning.

Who Is It For

For penetration testers, red teamers, security engineers, and AI/ML engineers who need hands-on offensive skills for LLM-powered applications. Also relevant for AppSec and DevSecOps professionals integrating AI into existing pipelines.

read less read more

What You’ll Learn

Exploit all 10 vulnerability classes in the OWASP® Top 10 for LLM Applications, including prompt injection, guard bypass, indirect injection, memory poisoning to multi-agent pipeline poisoning, and recommend LLM-native defensive architectures to remediate them.

read less read more

What It Prepares You For

Move into AI red-teaming, LLM penetration testing, and AI security engineering roles by demonstrating the ability to assess, exploit, and advise on the security of LLM-powered applications.

read less read more

Course Outline

Expand All

Collapse All

Introduction & Setup

LLM architectures (RAG, agents, multi-agent), OWASP® Top 10, attack surface mapping.

Direct Prompt Injection

RAG attacks, semantic retrieval, data extraction, dual-LLM defenses.

Guard Bypass

Encoding bypass, synonym attacks, multi-step extraction, guard hardening.

IP/Header Spoofing

X-Forwarded-For, LLM auth delegation risks, context injection.

Agent Tool Abuse

LangChain/LangGraph abuse, command execution, tool scoping.

Indirect Injection & SSRF

Data vs instruction boundary, webhook SSRF, sanitization.

Multi-Modal Injection

Vision model attacks, steganography, dual-vision guards.

Memory Poisoning

Conversational memory abuse, escalation spoofing.

Schema Confusion

Function-calling abuse, path traversal, tool ambiguity.

Customer Support AI

Multi-Agent Poisoning

Capstone & Wrap-Up

Prerequisites

Familiarity with web application security (HTTP, REST APIs, input validation, injection attacks).
Experience with an HTTP interception tool such as Burp Suite or OWASP® ZAP.
Python proficiency at a read-and-modify level, and the ability to interact with APIs using curl or Python requests.
Basic awareness of LLMs, system/user prompt structure, RAG as a concept, and LLM agents and tool calling at a conceptual level.

Lab Info