● HANDS-ON AI AppSec WORKSHOP
Red Teaming the Algorithm: Breaking and Securing Modern AI Systems
Move beyond basic prompt injection. Learn to identify, exploit, and mitigate critical vulnerabilities across the entire AI application lifecycle, from simple wrappers, RAG systems and autonomous agents to insecure output handling in an intensive 2-day hands-on AI security workshop.
This training is delivered through custom-built labs utilizing a dedicated lab environment to support intensive, hands-on exploitation of AI-integrated applications.
Across two intensive 5-hour sessions, you will adopt the attacker’s mindset to dismantle AI-driven architectures. You will work hands-on with real-world scenarios, exploiting Large Language Models (LLMs), bypassing safety filters, and hijacking autonomous agents to understand how to build truly resilient AI implementations.
● WHAT YOU’LL DO
What you’ll work on
LLM Exploitation & Jailbreaking
Master advanced prompt injection (Direct & Indirect) and jailbreaking techniques to bypass model alignment and safety guards.
RAG & Data Poisoning
Exploit Retrieval Augmented Generation (RAG) systems by poisoning data sources and manipulating the context injected into the LLM.
Agent Hijacking
Target autonomous agents and tool-calling interfaces to execute unauthorized actions and achieve remote code execution (RCE).
Insecure Output Handling
Demonstrates how unvalidated LLM outputs lead to traditional web vulnerabilities like XSS, CSRF, and SSRF in an AI context.
Supply Chain & Plugin Security
Analyze the risks of third-party model dependencies, insecure plugins, and the broader AI software supply chain.
Adversarial Defense & Hardening
Apply the OWASP Top 10 for LLMs framework to implement robust input/output filtering and secure architecture patterns.
● AGENDA
Your 2-day training journey
Day 1
Attacking the Model & Data Layers
→ Introduction to AI AppSec & the LLM Threat Landscape
→ Direct vs. Indirect Prompt Injection
→ Advanced Jailbreaking & Bypassing Safety Filters
→ Training Data & RAG Context Poisoning
→ Lab: Breaking Model Alignment & Exploiting Insecure RAG
Day 2
Exploiting Integrations & Autonomous Agents
→ Hijacking Autonomous Agents & Tool-Calling
→ Insecure Output Handling: AI-driven XSS and SSRF
→ Supply Chain Vulnerabilities & Insecure Plugins
→ OWASP Top 10 for LLMs: Defense-in-Depth for AI
→ Final Lab: Chaining AI exploits for full system compromise
● INSTRUCTORS
Meet the instructor
Spyros Gasteratos
Veteran security leader and Founder of Grafos AI, the safe and secure AI agent platform that automates Infrastructure as Code (IaC) editing for developers. A leading voice in Agentic AI Security and a key contributor to the OWASP AI Exchange, Spyros specializes in deploying autonomous systems that enhance developer velocity without compromising architectural integrity. With nearly two decades of experience ranging from hands-on engineering to the CISO’s office, he is a frequent speaker at global stages like DEF CON and QCon, where he translates complex security theory into pragmatic, automated workflows.
● WHAT YOU’LL GAIN
Key takeaways
→ Exploitation Mastery: Ability to perform complex multi-stage attacks on LLM-powered applications.
→ Architectural Insight: Deep understanding of how RAG and Agents introduce new attack vectors.
→ Strategic Defense: Practical skills to mitigate most attacks and build defence in depth auto-recovery and notification.
→ Adversarial Mindset: Expertise in red teaming stochastic systems to identify “unforeseeable” failure modes.
● REQUIREMENTS & PREREQUISITES
Requirements
→ Technical Proficiency: Basic understanding of Python and Web AppSec (OWASP Top 10) fundamentals.
→ Tooling: Familiarity with API interaction tools (e.g., Postman, cURL) and basic command-line usage.
● BOOK YOUR SEAT
Step into the action
Join us for two days of hands-on AI security exploitation.
Book your seat
What's included:
- 2 days (6h/day) of live expert-led instruction and real-world AI exploitation demos
- Exclusive lab access to hands-on environments specifically designed to host vulnerable LLM applications
- Certificate of Completion to validate your Practical AI AppSec skills
