AI Security: Principles & Practice

Practical theory and implementation for developers.

Learn theoretical and hands-on approaches to AI security and safety. This course covers:
- Foundations of machine learning
- Adversarial ML
- The cutting edge architecture of Transformers
- LLMs and red-teaming chat applications
- Differential Privacy

Sign up for a free account to track your progress.

We are grateful to Open Philanthropy for sponsoring the development and teaching of this free course.

Apply to a live session of this course

Course Introduction

Learn about the course, target audience and prerequisites.

Course Introduction

Introduction to ML

In this module, we will develop an intuitive understanding of of how computers learn from data. This will give us a good foundation for the rest of the course, and help us to understand some of the more cutting-edge model architectures used in text understanding and generation.

Adversarial ML

In the second module, we learn various adversarial methods to change the behaviour of models, and study how to approach red teaming and security for AI applications.

Transformers and LLMs

Now that we have covered the principles, practice and some of the security considerations of machine learning, we move onto the big league of large language models, or LLMs.

Red Teaming LLMs

We turn our attention to the unique security concerns of LLM-powered applications, and adopt the red-teamer mindset to consider and practice various attacks.

Differential Privacy

Learn about approaches to preserving statistical and useful information in datasets without comprising the privacy of the individuals whose data is included. We will attack and defend some approaches to DP.

Capstone

Build an AI security-related project to show new skills and explore research directions of interest.

Overview

Ready to start?

Select a lesson from the sidebar to begin.