Adversarial AI: Penetration Testing AI Systems
Testing an AI system is not like testing a web application, and this adversarial AI training in Toronto teaches the difference. You will learn a repeatable methodology for penetration testing models and the applications around them.
The work covers prompt injection and jailbreaks, training data extraction, and model theft, all inside the academy's isolated lab against systems you are authorized to test. You also learn to report findings that are probabilistic rather than a clean pass or fail.
- Duration
- 6 weeks
- Format
- Live online, instructor-led
- Prerequisites
- AI Security Principles and Practices: GenAI and LLM Defence
- Class size
- Capped at 12 learners
What you will be able to do
- Scope and plan a penetration test of an AI system
- Run direct and indirect prompt injection methodically
- Attempt training data extraction and model theft in the lab
- Test guardrails and measure how reliably they hold
- Rate findings that are probabilistic rather than binary
- Write an AI penetration test report a developer can act on
Course outline
5 modules
- Non-determinism and why one test is not enough
- The AI attack surface: model, prompt, data, tools
- Scoping and authorization for AI targets
- Setting up a safe testing harness
What you need before you start
- You have completed AI Security Principles and Practices or equivalent
- You have some penetration testing or application security background
- You can use an API and read simple scripts
- You accept that all testing is authorized and lab-confined
Who this course is for
- Penetration testers extending into AI systems
- Application security engineers whose products now include AI
- AI and ML engineers who want to test their own systems
- Security consultants asked to assess AI features
Questions about this course
More in ai security
CoreAI Security Principles and Practices: GenAI and LLM Defence
The foundational AI security course: threat model generative AI, defend against prompt injection, and set pragmatic policy for an enterprise rolling out AI.
8 weeks · live online
Contact for pricingView outline
CoreBuilding and Securing RAG and Agentic AI Systems
Build retrieval augmented generation and agentic AI hands on, then secure them: indirect injection, vector store isolation, and constrained agent tools.
8 weeks · live online
Contact for pricingView outline
CoreAI Powered Security Automation with Python
Automate security operations with Python, large language models, and the Model Context Protocol, building agentic tools that fail safely and keep a human in the loop.
8 weeks · live online
Contact for pricingView outline
Ready to launch your cyber security career?
Join the next live online cohort. No experience required, just bring the curiosity.