🎯 What You'll Learn
- Attack LLM applications end to end
- Poison data and extract models
- Craft adversarial evasion inputs
- Design layered AI defences
About This Workbook
A deep, hands-on follow-up to the AI Red Teaming workbook. This advanced track covers the full AI attack surface against realistic deployments: prompt and output attacks, data poisoning, adversarial evasion, model extraction, and the defences that counter them.
Chapter 1 — LLM Application Attacks
Exploiting the systems built around models.
- Direct and indirect prompt injection
- Insecure output handling
- Tool and plugin abuse
Chapter 2 — Data & Model Attacks
Targeting the pipeline and the model itself.
- Data poisoning
- Model extraction
- Membership inference
Chapter 3 — Adversarial Evasion
Making models see what is not there.
- Gradient-based attacks
- Sparse perturbations
- Robustness testing
Chapter 4 — Defending AI
Hardening models and applications.
- Input/output guardrails
- Adversarial training
- Monitoring and abuse detection
Keep going. Work through each chapter in order, then apply what you learned in the matching labs.