Adversarial AI: Attacking and Defending AI/ML Systems
Instructed by Charles Givre
Students learn how AI and ML systems fail before they learn to defend them. The course covers adversarial inputs, training pipeline compromise, model theft, and LLM-specific attacks. Labs target a multi-component AI stack from data ingestion through production inference.
Classes are limited. Get registered!
Learn the following key skills:
Course Outline
Day - How AI Systems Fail
Topics
- The AI attack surface: data, training, model, API, application
- AI Kill Chain threat model framework
- Attacker objectives: evasion, extraction, poisoning, inference, hijacking
- LABS:
- Threat model a production ML development
- Probe a black-box model to map decision boundaries
Day 2 - Training Pipeline Attacks
Topics:
Data poisoning and backdoor injection
Compromising ML infrastructure (MLflow, SageMaker, Kubeflow)
Model supply chain risk; SBOM/AIBOM and provenance
- LABS:
Inject a backdoor trigger into a training dataset
Exploit a misconfigured MLflow server; generate an AIBOM
Day 3 - Model Exploitation
Topics
- Adversarial examples against classifiers and security ML models
- Membership inference and model inversion attacks
- Model extraction via API queries
- LABS:
Craft adversarial examples evading a malware classifier
Model extraction and stolen model detection
Day 4 - LLM & GenAI Attack Surface
Topics:
Prompt injection (direct and indirect via documents)
System prompt extraction; jailbreak techniques
RAG pipeline attacks; agent tool abuse; endpoint misconfigs
- LABS:
Indirect prompt injection via poisoned RAG document
Agent tool abuse: exfiltrate data through a tool call
Day 5 - Defensive Engineering & MLSecOps
Topics:
Security gates in the ML CI/CD pipeline
Runtime monitoring: model behavior drift detection
AI incident response: containment, investigation, rollback
- LABS:
- Build a security-gated ML CI/CD pipeline
- Capstone: full-scope attack-and-defend exercise
Intermediate to Advanced
In person & Live Online
TBA
5 days, 40 hours
- Security engineers evaluating AI component risk
- ML engineers building security into pipelines
- Red teamers expanding into AI/ML targets
- Blue team leads building AI detection capabilities
Contact Us
Prerequisites
Python: read, modify, and run scripts
Familiarity with one cloud platform (AWS/Azure/GCP)
General application security background
No prior ML knowledge required
Why choose the Center for Cyber Security Training
Interactive, classroom-based learning
Subject matter experts
Trusted by US government agencies
Charles Givre
Charles Givre is a solutions-focused Senior Technical Executive and Data Scientist with more than 20 years of success across the technology, data science, Fintech, education, and cybersecurity industries. His broad areas of expertise include data visualization, analytics, network security, information assurance, strategic planning, big data, business intelligence, software development, and program management. He has held leadership positions at organizations including JP Morgan Chase, Deutsche Bank, Booz Allen Hamilton, and the Central Intelligence Agency (CIA). As VP and Product Owner for Data & Analytics at JP Morgan Chase, he led the product team to develop a next generation cybersecurity analytics platform. Mr. Givre is a published author, regular conference speaker and Project Management Committee (PMC) chair for the Apache Drill project.
The knowledge I gained through this course was substantial. From learning new script writing techniques to understanding big data tools, the instructor covered everything I was hoping for and more!- Sarah H., Washington, DC
Want more information?
Download the Applied Data Science for Cyber Security course outline now.
Upcoming Training Sessions
TBA - 2027
Classes are limited.
Enroll Now.
Related Courses
Our classroom delivers the most in-demand content from the highest profile subject matter experts. Intense and interactive, our courses prepare students with actionable insight and proven strategies.
Physical Penetration Testing
In this intense five-day course, students will learn a variety of covert techniques used to enter secure commercial facilities. From lock picking and key decoding to RFID cloning and alarm sensor bypassing, students who demonstrate hands-on proficiency in key tactics will earn a Certificate of Completion.
Exploit Development Bootcamp & Advanced
Our three-day Bootcamp will teach both basic & advanced techniques from a leading exploit developer. In our Advanced course, experienced students will learn how to write exploits that bypass modern memory protections for the Win32 platform in a fast-paced, interactive learning environment.
