Please note that this list of topics is based on our standard course offering, evolved from typical industry uses and trends. We will collaborate with you to tune this course and level of coverage to target the skills you need most. Topics, agenda and labs are subject to change, and may adjust during live delivery based on audience skill level, interests and participation.
Day 1: AI/ML Foundations and Attack Fundamentals
AI/ML Security Foundations
Understanding artificial intelligence and machine learning from a security perspective - establishing the essential knowledge base for identifying and defending against AI-specific threats.
- Overview of the OWASP Top 10 Application Security Vulnerabilities. Since AI models are frequently embedded within traditional web or enterprise applications, they inherit many of the same security risks identified by the OWASP Top 10. Understanding these common vulnerabilities is essential for developers and security professionals to protect both traditional and AI-powered applications from cyber threats.
- Essential AI/ML concepts for security professionals: supervised vs unsupervised learning, neural networks, deep learning fundamentals
- AI system architecture and deployment patterns: training vs inference, model serving, API endpoints
- The AI threat landscape: why traditional security approaches fail with AI systems
- Understanding the AI attack surface: training data, models, inference endpoints, and integration points
- Hands-on Lab (Jupyter Notebook): Setting up an AI security testing environment and exploring vulnerable ML models
Data Poisoning and Training Attacks
Deep dive into attacks targeting the AI training process, including practical implementation of data poisoning techniques and defense strategies.
- Data poisoning fundamentals: targeted vs untargeted attacks, clean-label attacks
- Training data vulnerabilities: data sources, collection pipelines, and validation gaps
- Backdoor attacks in machine learning models: trigger insertion and activation
- Supply chain security for AI: malicious datasets, compromised pre-trained models
- Hands-on Lab (Jupyter Notebook): Implementing data poisoning attacks against image classifiers and text models
- Hands-on Lab (Jupyter Notebook): Building data validation pipelines and poisoning detection systems
Day 2: Adversarial Attacks and Model Security
Adversarial Examples and Model Manipulation
Comprehensive exploration of adversarial attacks against deployed AI models, including hands-on generation of adversarial examples and evasion techniques.
- Adversarial examples: perturbation-based attacks, gradient-based methods (FGSM, PGD)
- Model evasion techniques: black-box vs white-box attacks, query-based optimization
- Physical world adversarial attacks: adversarial patches, real-world evasion
- Transferability of adversarial examples across different models and architectures
- Hands-on Lab (Jupyter Notebook): Generating adversarial examples using popular attack frameworks
- Hands-on Lab (Jupyter Notebook): Testing adversarial robustness of production AI systems
Privacy Attacks and Information Extraction
Understanding how attackers can extract sensitive information from AI models, including membership inference and model inversion attacks.
- Membership inference attacks: determining if specific data was used in training
- Model inversion attacks: reconstructing training data from model parameters
- Property inference: extracting global properties about training datasets
- Model extraction and stealing: replicating proprietary models through queries
- Hands-on Lab (Jupyter Notebook): Conducting membership inference attacks against machine learning models
- Hands-on Lab (Jupyter Notebook): Implementing model inversion techniques to extract sensitive information
- Differential privacy fundamentals and implementation strategies for AI systems
Day 3: Secure AI Integration and Enterprise Defense
Securing AI-Integrated Applications
Practical security implementation for traditional applications that leverage AI models and services, including LLM integration patterns.
- Secure API integration patterns for AI services: authentication, rate limiting, input validation
- LLM integration security: prompt injection attacks, output validation, context isolation
- Building secure AI microservices: containerization, network isolation, monitoring
- Input sanitization for AI systems: handling untrusted data, format validation
- Hands-on Lab: Implementing secure LLM integration using the Hugging Face Inference API (Python/Flask, Java/Spring, ASP.Net, Node.js options)
- Hands-on Lab: Building input validation pipelines for AI-powered web applications in your chosen language
Enterprise AI Security Strategy
Comprehensive approach to building organizational AI security capabilities, including governance, monitoring, and incident response.
- AI security governance frameworks: risk assessment, policy development, compliance
- Continuous monitoring for AI systems: model drift detection, anomaly identification
- AI security testing and red teaming: automated testing, adversarial validation
- Incident response for AI breaches: containment strategies, forensic analysis
- Hands-on Lab: Setting up AI security monitoring dashboards and alerting systems
- Hands-on Lab: Conducting AI security assessments and building remediation plans
Advanced Topics and Emerging Threats
Exploration of cutting-edge AI security challenges and future threat vectors.
- Large Language Model (LLM) specific attacks: jailbreaking, instruction following exploits
- Multi-modal AI security challenges: vision-language models, cross-modal attacks
-
- AI supply chain security: model provenance, dependency management
- Regulatory compliance for AI systems: GDPR, algorithmic auditing requirements
Course Wrap-up and Resources
- Next steps in your AI security journey
- Essential tools and frameworks for ongoing AI security work
- Building and maintaining AI security expertise within your organization
- Community resources and continued learning opportunities
