AI Security Masterclass: Prompt Injection & LLM Security

Build AI applications that are not only powerful, but secure by design.

In this hands-on AI Security Masterclass, you will learn how to build, attack, and secure modern applications powered by Large Language Models (LLMs). Using Python and Ollama, you will create a complete AI assistant and progressively add chat, Retrieval-Augmented Generation (RAG), tool calling, persistent memory, autonomous agents, and production security controls.

You will begin by building a local AI chat assistant and exploring why LLM security differs from traditional application security. You will then launch realistic prompt injection attacks, attempt to expose hidden instructions, test role-playing and multi-turn jailbreaks, and experiment with encoded prompt manipulation techniques.

Next, you will build a document-based RAG application capable of answering questions from PDFs. You will see how malicious documents can introduce indirect prompt injection, poison retrieved context, manipulate answers, and undermine trusted knowledge. You will secure the RAG pipeline using source validation, context isolation, sanitization, and enforcement controls.

The course then moves beyond chatbot responses into real application actions. You will build a tool-calling AI assistant with weather, calculator, and email tools. You will test parameter injection and unauthorized tool execution before implementing input validation, least-privilege permissions, destination restrictions, and human-in-the-loop approval for sensitive actions.

You will also add persistent AI memory and learn how attackers can poison stored information to manipulate future conversations. You will protect memory using validation, approval workflows, risk scoring, and safe storage policies.

As the assistant becomes more autonomous, you will build an AI agent capable of planning tasks, selecting tools, and executing multi-step workflows. You will then attempt to compromise its decisions and secure it with permissions, policy enforcement, approval gates, and constrained execution.

By the end of the course, you will combine every defense into a reusable AI Security Gateway that includes:

• Prompt injection detection

• Jailbreak filtering and risk scoring

• RAG security and context sanitization

• Tool authorization and parameter validation

• Memory poisoning protection

• Agent security controls

• Human approval workflows

• Output validation and audit logging

Finally, you will package and run the complete secure AI assistant using Docker and connect it to local models through Ollama.

This course is ideal for Python developers, Generative AI engineers, RAG developers, AI agent builders, software engineers, and technical professionals who want practical experience securing real LLM applications.

You will not simply study AI attacks. You will build vulnerable features, exploit them, implement defenses, and verify that those defenses work.

Build it. Attack it. Secure it.