Discover how attackers corrupt the data AI systems rely on, from RAG pipelines to training sets.

This module descends into the deepest layer of AI security: the data. Learners explore risks specific to Retrieval-Augmented Generation (RAG) systems, attacks targeting ingestion pipelines and embeddings, and how private data can be exposed through retrieval or embedding leakage. Mapped to OWASP LLM02, LLM04, and LLM08, the module closes with two scenario-based challenges centred on RAG poisoning detection and defence.

RAG Security Fundamentals

Learn how RAG systems work and how attackers exploit retrieval, context, and trust boundaries.

Data Poisoning in RAG Systems

Explore how data poisoning alters AI embeddings and retrieval results without visible errors.

Sensitive Information Disclosure

Explore how AI embeddings, retrieval, and weak access controls expose sensitive private data.

UnIndexed

An AI assistant was given access to everything. Nobody checked what "everything" included.

Lockdown

An AI assistant has three open vulnerabilities. Find them. Fix them. Prove the system is secure.

Topic Rewind Recap

Lock in what you learned with a recap. Earn points and keep your streak.

What are modules?

A learning pathway is made up of modules, and a module is made of bite-sized rooms (think of a room like a mini security lab).

Hierarchical diagram showing how learning pathways contain modules, which contain individual rooms.