Free Course · 5 lessons · ~3.5 hours
AI Security Foundations
What every security professional needs to know about AI systems. The new attack surface, the threat models that work, the controls that matter, and the regulatory frame you have to operate inside.
Who this is for
You should take this course if you are responsible for the security of an organisation that uses, builds, or buys AI systems — and you would like to stop relying on vendor marketing for your view of what AI security actually is.
The audience is wider than "AI engineers". You should take it if you are a CISO, a security architect, a compliance lead, a GRC professional, a security engineer, an auditor, a risk officer, or a board member who has to ask the right questions. The course assumes you are familiar with normal information security (firewalls, identity, vulnerability management) and adds the AI-specific layer on top.
If you are looking for a tutorial on building secure AI applications from scratch, this is the wrong course. That is what Path D of the Integrated AI Program is for. This course is the orientation that comes before the building.
What you will learn
Why AI is a new security domain
What is genuinely different about securing AI systems compared with traditional software. The new attack surface, the new blast radius, the frameworks that already exist.
The OWASP Top 10 for LLMs
The ten most common vulnerabilities in language-model applications, walked through with examples and clear language. The vulnerabilities, in 2026, are well documented; the question is whether you recognise them in your own systems.
Threat modelling with MITRE ATLAS
The threat-modelling framework specifically for machine-learning systems. How it relates to MITRE ATT&CK. A worked threat model for a realistic system.
Defences that work
Defence-in-depth for AI. Input validation, output filtering, tool allowlists, monitoring, red-teaming. The MLSecOps mindset. What you can deploy this quarter; what is harder.
The compliance and regulatory frame
EU AI Act, NIS2, DORA, ISO 42001, NIST AI RMF. What each one requires, where they overlap, and how to map your AI systems to them. You finish the course with a one-page AI security posture document.
What you will need
- About three and a half hours of total time. Each lesson is 30 to 50 minutes.
- A working knowledge of traditional information security. You do not need to be a developer.
- For the threat modelling exercise: paper or a whiteboarding tool you are comfortable with.
- No installations. No code. No accounts beyond your normal ones.
How this fits with the full programme
This free course is a serious orientation, not a substitute for the full discipline. The Integrated AI Program's Path D — AI for Cybersecurity is sixty ECTS of dedicated training across ten courses, including adversarial machine learning, MLSecOps, AI governance and compliance for security, privacy-preserving AI, and AI-augmented incident response. The wrap-up page at the end of this course lists the specific Path D modules that pick up where the free material leaves off.