AI Security: Terminating The Terminator
Training Overview
AI Security: Terminating the Terminator is a hands-on, concept-to-attack training at CSA XCON 2026, designed to help security professionals, developers, and AI practitioners understand how AI systems work and how they fail under attack.
As AI becomes deeply embedded into applications, security controls, and decision-making systems, it introduces entirely new attack surfaces. From adversarial inputs and data poisoning to model theft, prompt injection, and LLM jailbreaks this training explores how attackers abuse AI systems and how defenders can stop them.
The course combines AI fundamentals, real-world case studies, and carefully crafted labs to ensure participants gain both intuition and practical skills to secure modern AI and GenAI systems.
About the Training at CSA XCON 2026
This is not just an AI course, and not just a security course it sits precisely at the intersection of machine learning, GenAI, and offensive/defensive security.
Key highlights include:
- Understanding how AI models are built and deployed
- Hands-on exploration of ML and GenAI vulnerabilities
- Attacks against computer vision, LLMs, RAG, and agentic AI
- Practical mitigation strategies and security frameworks
- Real-world examples of AI systems failing under attack
This training aligns with CSA XCON’s mission to prepare professionals for next-generation cyber threats.
What You Will Learn
1 – How AI Systems Work
Goal: Build intuition about AI architectures before attacking them.
Foundations of AI
- Core AI and ML terminology
- Overview of common AI frameworks
- Understanding supervised and unsupervised learning
Neural Networks & Deep Learning
- How neural networks function
- Classification models (spam filters, WAFs, fraud detection)
- Convolutional Neural Networks (CNNs) for: Image classification, Object detection & Face recognition
Generative AI & LLMs
- Why GenAI is different
- Transformer architecture & attention mechanisms
- Large Language Models (LLMs)
- Retrieval-Augmented Generation (RAG) systems
- Agentic AI and Model Control Protocols (MCPs)
Deployment & MLOps
- End-to-end AI pipelines
- Model training, serving, and monitoring
- Introduction to MLOps risks
2 – Breaking AI Systems
Goal: Identify and exploit vulnerabilities unique to AI and GenAI.
Core AI Security Vulnerabilities
- Adversarial learning attacks
- Fooling image classifiers and object detectors
- Physical-world adversarial patches
- Attacks against face recognition systems
Model Attacks
- Model stealing and extraction
- Model inversion and privacy leakage
- Model skewing and data poisoning
- Attacking feedback loops in AI systems
GenAI & LLM Attacks
- Prompt injection and jailbreak attacks
- Owning LLM behavior
- Attacking RAG pipelines
- Risks in internal chatbots and enterprise AI tools
Agentic AI & MLOps Risks
- Breaking agentic AI workflows
- What can go wrong with MCP-based systems
- Insecure serialization of ML models
- Vulnerabilities in MLOps frameworks
Security Frameworks for AI
- MITRE AI threat modeling
- OWASP Top 10 for AI & LLMs
Training Experience & Expectations
- Hands-on labs throughout the training
- Beginner-friendly AI explanations with security depth
- Real-world examples of AI systems under attack
- Clear mapping between vulnerabilities and mitigations
Participants will leave with actionable AI security knowledge, not just concepts.
Who Should Attend
This training is ideal for:
- Cybersecurity professionals securing AI systems
- Application & ML developers
- Blue team and architecture teams
- Students with CS/AI background
- AI and GenAI enthusiasts
Skill Level
Beginner to Intermediate
Participant Requirements
Participants should bring:
- Laptop with internet access and ability to expose a public HTTP endpoint
- Gmail account (for Google Colab labs)
- Basic Linux command-line familiarity
- Curiosity and readiness for hands-on experimentation
Basic Python or ML knowledge is helpful but not mandatory.
What Participants Will Receive
Each participant will receive:
- Course slides and structured notes
- Pre-built lab exercises for AI and GenAI security
- Post-training reference material
- A roadmap for continuing AI security learning
What This Training Is
What not to expect:
- Clear understanding of AI systems
- Practical exposure to AI attack techniques
- Mitigation strategies grounded in real scenarios
What not to expect:
- Becoming an AI security expert
- Heavy mathematical derivations