AI Security: Preventing Prompt Injection
Course Overview
Advanced
Free Course
For engineers who ship LLM features and agents that read untrusted text or call tools. You will learn how prompt injection and jailbreaks work, how to red-team and evaluate an LLM application, and how to build the deterministic guardrails — scoped tools, policy checks, egress rules and monitoring — that keep a fooled model from doing harm.
Instructor: Jaidev
Sections: 3
Course Content
Section 1: Prompt Injection and Jailbreaks
How injected instructions reach a model, how to test an LLM application for them, and which defences still hold when the model is fooled.
Section 2: AI Safety and Responsible Guardrails
How to filter harmful content, enforce policy on every tool call in code, and measure whether your safety controls actually work.
Section 3: Mini Project
Build and test a layered safety wrapper for an LLM support assistant, and prove each layer catches something the others miss.