Adversarial ML & Security Threats

• Chapter 1.1: What is Adversarial ML?

• Chapter 1.2: Historical Context and Emerging Importance

• Chapter 1.3: Types of Adversarial Threats (White-box, Black-box, Gray-box)

• Chapter 1.4: Overview of Vulnerabilities in ML Pipelines

• Chapter 2.1: Evasion Attacks on Image, Text, and Tabular Models

• Chapter 2.2: Poisoning Attacks During Training

• Chapter 2.3: Model Inversion and Membership Inference

• Chapter 2.4: Tools and Libraries (Foolbox, ART, CleverHans)

• Chapter 3.1: Adversarial Training Techniques

• Chapter 3.2: Input Preprocessing and Gradient Masking

• Chapter 3.3: Certified Defenses and Formal Guarantees

• Chapter 3.4: Evaluation Metrics for Robustness

• Chapter 4.1: Secure Data Pipelines and Label Integrity

• Chapter 4.2: Attack Surface in Model Deployment

• Chapter 4.3: Threat Modeling for ML Systems

• Chapter 4.4: Secure MLOps and Monitoring Pipelines

• Chapter 5.1: Case Studies: Attacks on Facial Recognition, NLP, and Healthcare Models

• Chapter 5.2: Adversarial Threats in Federated Learning and Edge AI

• Chapter 5.3: Legal, Ethical, and Compliance Risks

• Chapter 5.4: AI Red Teaming and Offensive Testing

• Chapter 6.1: Design Your Own Adversarial Attack Scenario

• Chapter 6.2: Simulate and Evaluate Defense Mechanisms

• Chapter 6.3: Final Capstone Project Presentation

• Chapter 6.4: Future Directions – AI Security, Regulation, and Red-Blue Team Dynamics

~Video content aligned with weekly modules

Theme: Foundations and Attack Vectors in Adversarial ML

• What is Adversarial Machine Learning?

• Threat Models: White-box, Black-box, and Gray-box Attacks

• Vulnerabilities in ML Pipelines and Workflows

• Evasion Attacks: Images, Text, and Structured Data

• Data Poisoning Attacks During Model Training

• Membership Inference and Model Inversion Attacks

• Tools of the Trade: CleverHans, Foolbox, ART

• Case Demo: Crafting and Launching an Evasion Attack

Theme: Defense Mechanisms and Secure ML Pipelines

• Adversarial Training: Concepts and Implementation

• Gradient Masking, Preprocessing, and Randomization Defenses

• Certified Defenses and Robustness Verification

• Evaluating Model Robustness: Metrics and Benchmarks

• Securing the ML Lifecycle: Data Collection to Deployment

• Threat Modeling for AI Systems

• Safe Deployment: API Security and Model Hardening

Theme: Real-World Applications and Red Teaming Practice

• Adversarial ML in Biometric Systems and NLP Models

• Attacks in Federated Learning and Edge AI Environments

• AI Red Teaming: Tools, Strategies, and Testing Pipelines

• Ethics and Compliance in Adversarial Research

• Simulating an End-to-End Attack and Defense Workflow

• Real-World Incident: AI Attack Forensics Walkthrough

• Capstone Challenge: Attack Design and Mitigation Exercise

• Final Recap: Preparing AI Systems for Real-World Threats

Title: Understanding the Enemy: How Adversaries Exploit ML Systems
Duration: 60 minutes
Focus: Breakdown of major adversarial attack types and why current ML systems are vulnerable
Guest: Security Researcher or Adversarial ML Specialist
Interactive: Live threat modeling of a facial recognition or fraud detection system

Title: Defending the Pipeline: Engineering Robust and Secure ML Models
Duration: 75 minutes
Focus: Exploration of practical defenses, including adversarial training, gradient masking, and pipeline hardening
Guest: Applied ML Security Engineer / Secure MLOps Architect
Interactive: Guided demo of evaluating model robustness under adversarial input

Title: From Labs to the Real World: Red Teaming AI Systems at Scale
Duration: 90 minutes
Focus: Real-world implementation of offensive testing, red teaming workflows, and mitigation planning
Guest Panel: AI Red Team Lead + Threat Intelligence Analyst + Legal/Ethics Advisor
Interactive: Live walk-through of a full red team test plan with audience Q&A on incident response

Edit

Edit