AI & LLM Security (OWASP LLM Top 10)

What you will be able to do

• Explain the OWASP Top 10 for LLM Applications and what each category actually means
• Identify direct and indirect prompt injection in real application flows and design defenses against both
• Recognize sensitive-information-disclosure and training-data risks and apply data-minimization and output controls
• Treat LLM output as untrusted and apply output handling, sandboxing, and least-privilege to curb excessive agency
• Assess supply-chain and model risks across base models, fine-tunes, plugins, and datasets
• Plan and run AI red-teaming and adversarial testing, and select defensive controls that map to specific risks
• Map LLM-specific threats to a layered control architecture and communicate residual risk to stakeholders

What is inside

4 modules, 8 lessons. Each module ends in a graded quiz and most carry an assignment.

1. 01

   The Threat Landscape & the OWASP LLM Top 10

   Before you can defend an LLM system you have to know what it actually is and where it can be attacked. This module builds a mental model of how an LLM application works, why it widens the classic attack surface, and then walks the OWASP Top 10 for LLM Applications so you can name and recognize every category.

   2 lessons · 5 quiz questions

2. 02

   Prompt Injection, Direct & Indirect

   Prompt injection is the signature LLM vulnerability and LLM01 on the OWASP list. This module separates the two main forms, direct and indirect, shows realistic attacks against an assistant that reads external data and uses tools, and then builds the layered defenses that actually reduce the risk, because there is no single complete fix.

   2 lessons · 5 quiz questions · assignment

3. 03

   Sensitive Data, Output Handling & Excessive Agency

   This module covers the risks on the data and output side of the pipeline: how LLM systems leak sensitive and training data (LLM02), why model output must be treated as untrusted before any downstream system uses it (LLM05), and how excessive agency (LLM06) turns a manipulated model into real-world harm. You will learn the data-minimization and output-handling controls that contain all three.

   2 lessons · 5 quiz questions

4. 04

   Supply Chain, Model Risk, Testing & Red-Teaming

   The final module covers risks you inherit rather than write (supply chain and model poisoning, LLM03/LLM04) and the practices that prove your defenses work: AI red-teaming, adversarial testing, and a layered defensive architecture. You will leave able to assess third-party model risk and plan a testing program, then put it all together in the capstone assignment.

   2 lessons · 5 quiz questions · assignment