AI Security
AI Security Fundamentals: Protecting AI Systems
Understand how AI systems fail, and how to defend them.
- 10 modules
- 16 labs
- ~20 hours
- Intermediate
- Hands-on Labs
- Projects
- Certificate
Course overview
AI systems introduce a new attack surface: prompts, training data, embeddings, tools and agents. This course teaches how each layer breaks and what controls actually reduce risk.
You test a real LLM application against OWASP's LLM risk categories, map findings to MITRE ATLAS, and produce a security review an engineering team can act on.
What you will be able to do
- Explain the architecture of modern AI and LLM applications
- Test for prompt injection, jailbreaks and data disclosure
- Assess RAG pipelines and vector store exposure
- Threat model an AI feature using ATLAS and the NIST AI RMF
- Recommend concrete guardrails and monitoring
Curriculum
- 1
AI & ML Foundations for Security
Enough model mechanics to reason about risk.
- Model types and training
- Inference pipelines
- Where trust boundaries sit
- 2
LLM Application Architecture
Prompts, context, tools and memory.
- System prompts
- Context assembly
- Tool calling
- Session memory risks
- 3
Prompt Injection & Jailbreaking
The dominant real-world LLM attack class.
- Direct injection
- Indirect injection via content
- System prompt leakage
- Mitigation patterns
- 4
RAG & Data Security
Retrieval is an attack surface.
- Vector store access control
- Knowledge base poisoning
- Sensitive data disclosure
- Chunk-level authorisation
- 5
Agents, Tools & Excessive Agency
When the model can act, blast radius matters.
- Agent architectures
- Tool abuse
- MCP security considerations
- Human-in-the-loop design
- 6
AI Threat Modeling & Governance
Turn findings into a defensible programme.
- MITRE ATLAS mapping
- NIST AI RMF
- AI incident response
- Responsible AI controls
Labs you will build
PrimeSec does not hand you a pre-built machine. You get professional lab guides and build the environment yourself — that is where the skill comes from.
- Stand up a local LLM application to test against
- Execute and document a direct prompt-injection chain
- Trigger an indirect injection through retrieved content
- Extract a system prompt and design a mitigation
- Poison a RAG knowledge base and detect it
- Constrain an agent's tool permissions
- Build an AI threat model for a real feature
Portfolio projects
- A full AI security assessment report with severity ratings
- An AI threat model mapped to MITRE ATLAS
- A guardrail and monitoring recommendation pack
Frequently asked questions
Do you provide the lab environment?
No — and that is intentional. PrimeSec gives you professional lab guides that teach you to build and configure the environment yourself using your own machine, Hyper-V, VMware, VirtualBox, Docker, or a cloud free tier. Building and troubleshooting the environment is part of the skill.
Is this course self-paced?
AI Security Fundamentals is self-paced. Lessons, knowledge checks, labs and projects unlock in order so you always know what to do next.
Do I get a certificate?
You receive a PrimeSec course completion certificate once every module, lab and project requirement is met. It demonstrates completion and practical work — not an accredited industry certification.
Will cloud labs cost me money?
Labs are designed around free tiers and local virtualisation wherever possible, and every cloud lab includes cleanup steps so you do not leave billable resources running.
Related courses
Cloud Security
Cloud Security Engineer Bootcamp
Intermediate · 14 modules · 22 labs
IAM & Identity
Entra ID & Identity Security Mastery
Intermediate · 9 modules · 14 labs
DevSecOps
DevSecOps Engineering: Secure CI/CD Pipelines
Intermediate · 12 modules · 18 labs