Generative AI Risks & Cybersecurity: LLM Security

This course contains the use of artificial intelligence.

Generative AI is now inside customer chatbots, code assistants and document tools, and attackers have noticed. In this course you attack and then defend real AI systems, not just read about them.

What to expect: About 5 hours of video. Plan 1 hour to install the lab tools (Ollama and a local model), then about 9–12 hours to complete the 14 labs and their assignments. All 14 labs run in a single lab file on your own computer, using a free local Llama model through Ollama, so no API key or paid account is needed. Each lab has an optional assignment with a worked model answer; assignments are practice and are not required for your certificate.

What you will do in the labs:

• Extract a secret from a chatbot's system prompt with prompt injection, and see why "never reveal" instructions fail.
• Hide instructions in a document and watch an AI summarizer obey them (indirect prompt injection).
• Compare jailbreak techniques, probe hallucinations, and test a model for bias with paired prompts.
• Build guardrails and run a 10-attack battery against them, then add output validation and a content classifier you can score for accuracy.
• Classify an AI system under the EU AI Act, generate an AI use policy, respond to an AI security incident, and run a full red team exercise.

How the course is organized: nine sections take you from how generative AI works and the threat landscape, through prompt injection and jailbreaking, AI-specific attacks (data poisoning, model bias, SSRF through function calling, cost-based denial of service, hallucinations), and defense-in-depth, to secure development, red teaming, governance and compliance (EU AI Act, NIST AI RMF, ISO 42001), and emerging threats such as agentic AI. A running case study, NovaStar AI, ties every section to one realistic company.

What you need: a computer that can run Ollama (Windows, macOS or Linux), a modern browser, and basic familiarity with security concepts. No programming or machine learning background is required.

Who this course is for: security professionals, penetration testers, developers building AI features, and risk and compliance staff who need to understand how AI systems are attacked and how to protect them.