You can hold the AI security conversation.

What you have learned

• The five things that are genuinely new about securing AI systems — and the trap of treating the rest as new when it isn't.
• The 2025 OWASP Top 10 for LLM Applications, applied to a real system in a worked example.
• How to threat-model an AI system using MITRE ATLAS, walking through seven core tactics.
• The seven layers of defence in depth for AI — identity, input boundary, prompt, tool gating, output handling, monitoring, red-teaming — and what populates each.
• The five frameworks that matter: EU AI Act, NIS2, DORA, ISO 42001, NIST AI RMF — when each applies, where they overlap, and which to pick as your spine.
• A one-page AI security posture document you can refine and use.

If you finished this honestly, you now have more working knowledge of AI security than most CISOs do today. That is not a high bar — but it is the bar that matters when your audit committee asks you the question.

What this course did not cover

• The deep technical mechanics. How an adversarial example is actually constructed, how a prompt injection is technically defended at the model layer, how MLSecOps tooling integrates with ML pipelines. The maths and the engineering.
• Specific tool deployments. Configuring PyRIT, building a guardrail with Lakera, integrating Garak with CI. These are training-day topics, not free-course topics.
• Incident response specifics. The runbook for an AI-related breach, the forensics, the disclosure obligations across jurisdictions. Substantial body of work.
• Privacy-preserving AI. Federated learning, differential privacy, homomorphic encryption for AI workloads — a discipline of its own.
• Securing AI training pipelines. If you build or fine-tune models, the supply-chain and training-infrastructure security work is much larger than what we covered.

Where Path D takes this further

The Integrated AI Program's Path D — AI for Cybersecurity is sixty ECTS over ten courses. The map from this free course to the paid programme is roughly:

• D1: Cybersecurity Foundations for AI Practitioners. The deeper version of Lesson 1.
• D4: Adversarial Machine Learning and AI Red Teaming. The deeper version of Lesson 3, with hands-on attack construction.
• D5: Securing AI Systems (MLSecOps, Model Supply Chain, Prompt Injection Defence). The deeper version of Lesson 4.
• D7: AI Governance and Compliance for Security. The deeper version of Lesson 5, covering DORA, NIS2, ISO 27001, NIST CSF, and ISO 42001 in working detail.
• D8: Privacy-Preserving AI and Data Protection. The discipline we deliberately did not cover.
• D9: AI-Augmented Incident Response and Forensics. Operating in incident response when AI is both the target and a tool.

Path D launches with the December 2026 cohort. Applications are open from November 2026; scholarship places exist for a small percentage of seats.

Stay in touch

If you found this useful, we would love to know who you are. The form below sends a short note to our admissions inbox.

First name

Email

Your role CISO or security leader Security architect Security engineer Compliance / GRC professional Auditor (internal or external) Board member or executive Engineer interested in AI security Other Type of organisation (optional) Financial services Healthcare Public sector or critical infrastructure Technology vendor Professional services / consulting Other regulated industry Other Anything you want to tell us? (Optional)

Send Or apply now →

We will only use this to reply to you. We will not subscribe you to a newsletter without asking first.

Three immediate next steps

1. Share the course. If a colleague would find it useful, send them the course link.
2. Take our other free courses. Build Your First AI Agent covers the general agent-building patterns; Build a Market Research Bot applies AI to a regulated, education-only domain.
3. Apply for the December 2026 cohort. If you have read this far, you might be exactly the kind of learner Path D is built for. The application form takes ten minutes.