AI Security Engineering Foundations

What you will be able to do

• You will be able to map security requirements across every stage of the machine learning lifecycle, from data collection through training, deployment, and monitoring
• You will be able to explain how AI systems expand the attack surface beyond traditional software, and place each risk on a recognized taxonomy (NIST AI 100-2, MITRE ATLAS, and the OWASP LLM Top 10)
• You will be able to threat model an AI, LLM, RAG, or agent architecture by drawing its data flows and trust boundaries and enumerating the AI-specific threats that cross them
• You will be able to describe the four adversarial machine learning attack classes (evasion, poisoning, privacy, and model extraction) and reason about the defenses each one calls for
• You will be able to explain prompt injection, jailbreaks, and insecure output handling in LLM and agent systems, and design guardrails that reduce their blast radius
• You will be able to secure the AI supply chain by validating dataset provenance, verifying and signing model artifacts, and producing a machine learning bill of materials
• You will be able to apply secure model development practices across training, evaluation, and deployment, including AI red-teaming, least-privilege tool and agent design, and MLSecOps pipeline gates
• You will be able to build monitoring and detection for AI systems that maps observed behavior back to MITRE ATLAS tactics

What is inside

4 modules, 8 lessons. Each module ends in a graded quiz and most carry an assignment.

1. 01

   The Secure Machine Learning Lifecycle

   Before you can secure AI, you have to see the whole system it lives in. This module walks the end to end machine learning lifecycle, from data collection and training through deployment, monitoring, and retirement, and shows where security has to be engineered in at each stage. You will learn why AI systems break the assumptions traditional security depends on, and you will meet the standards that frame the field: the NIST AI Risk Management Framework, MITRE ATLAS, and the NIST AI 100-2 adversarial taxonomy.

   2 lessons · 5 quiz questions

2. 02

   Threat Modeling AI and LLM Architectures

   Threat modeling is the core skill of an AI Security Engineer, and AI systems need their own version of it. This module teaches you to diagram the four architectures you will meet in the field (predictive machine learning, LLM applications, retrieval augmented generation, and tool-using agents), mark the trust boundaries that classic web diagrams miss, and trace where data and model artifacts flow across each system. You then enumerate and rank AI-specific threats with three complementary frameworks: MITRE ATLAS for adversary tactics and techniques, the OWASP Top 10 for LLM Applications (2025) as a fast threat checklist, and an adapted STRIDE for category coverage. The module closes with a realistic assignment: build a full threat model, including a ranked threat register, for a customer-facing assistant that mixes retrieval with money-moving agent tools.

   2 lessons · 5 quiz questions · assignment

3. 03

   Adversarial Machine Learning and Model Threats

   Before you can defend a model you have to understand how it is attacked. This module works through the four adversarial attack classes named in NIST AI 100-2: evasion with adversarial examples, data and model poisoning including backdoors, privacy attacks such as membership inference and model inversion, and model extraction. It carries each class into the LLM layer (jailbreaks, prompt injection, and training-data extraction) and pairs every attack with the defense it calls for and that defense's real limits.

   2 lessons · 5 quiz questions

4. 04

   Secure Model Development Practices

   This is where the course turns into hands-on engineering. You will secure the AI supply chain by validating dataset provenance, verifying and signing model artifacts, and producing a machine learning bill of materials. Then you will harden training, evaluation, and deployment with AI red-teaming, input and output validation, and least-privilege design for tools and agents. Finally you will wire it all into MLSecOps: automated security gates in the CI/CD pipeline and production monitoring that maps live behavior back to MITRE ATLAS.

   2 lessons · 5 quiz questions · assignment