Practical AI Security: Attacks, Defenses, and Applications

Practical AI Security: Attacks, Defenses, and Applications is a comprehensive course built for developers and cybersecurity professionals who want to get up to speed on the fast-moving world of AI and security. Instead of staying high-level, it takes you step by step from the basics of machine learning to advanced offensive and defensive techniques involving Large Language Models (LLMs). By the end, you’ll know how to use AI as part of your security toolkit, while also spotting and defending against the new risks it brings.

The course starts with the essentials of AI and machine learning. You’ll learn the differences between AI, Machine Learning, and Deep Learning, and then actually train models yourself. With practical labs using tools like Scikit-Learn, you’ll get comfortable with concepts like supervised vs. unsupervised learning, neural networks, bias, and overfitting. This way you’ll know how AI works under the hood before jumping into security-specific applications.

From there, we’ll dig into how modern generative AI systems are built, focusing on LLMs. You’ll learn the core ideas behind tokenization, context windows, and the Transformer architecture that drives models like GPT. Through exercises, you’ll practice prompt engineering, tweak parameters like temperature and top-p to control outputs, and use embeddings with vector databases such as FAISS for semantic search. You’ll even get to build and deploy simple LLM-based web apps.

Once you’ve got the fundamentals down, we’ll move onto some real security applications. On the offensive side, you’ll explore how AI can be used for things like automated pentesting agents, vulnerability discovery, and exploit development. We’ll look at research projects like Google’s Project Naptime and review tools already used by red teams. On the defensive side, you’ll see how AI can help with threat modeling, automated code reviews, and security-focused retrieval systems for documentation and CVE databases.

Finally, the course zeroes in on the security of AI systems themselves. You’ll experiment with prompt injection attacks, denial-of-service techniques, and data exfiltration against AI agents, looking at real-world issues in systems like LangChain, GitHub Copilot, and ChatGPT. Then, you’ll learn how to defend against these attacks by building guardrails, applying frameworks like Google’s SAIF, and adapting red teaming approaches to generative AI.

By attending this course , you will get 

• Certificate of completion for the Training program
• Cloud Access for hands-on labs and exercises
• A curated list of essential articles and research papers in AI security
• Source code for all labs and custom tools
• Slack access for the class and after for regular AI security discussions