AI Red Teaming & LLM Security — OWASP LLM Top 10 & MITRE ATLAS
The fastest-growing cybersecurity specialty in 2026 — attack and defend AI systems before adversaries do.
About this course
AI Red Teaming is the hottest new cybersecurity specialisation in 2026. As enterprises deploy LLM-powered chatbots, agents and code generators, adversaries are actively targeting them through prompt injection, jailbreaks, model extraction and data poisoning. This course teaches you to systematically attack and defend AI systems using OWASP LLM Top 10, MITRE ATLAS, and industry-standard red teaming frameworks (Garak, PyRIT). Bridges your existing security or prompt engineering knowledge into the AI security domain.
What you'll achieve
- Execute prompt injection attacks — direct, indirect and multi-turn jailbreaks
- Apply OWASP LLM Top 10 (2025) across real AI application assessments
- Map AI attack chains using MITRE ATLAS threat framework
- Use Garak and PyRIT (Microsoft) for automated LLM red teaming
- Identify and exploit insecure agentic workflows and MCP tool abuse
- Implement LLM guardrails, input validation and output filtering
- Conduct formal AI security assessments and produce audit-ready reports
Curriculum
Module 1
AI Threat Landscape & MITRE ATLAS
AI attack taxonomy · MITRE ATLAS framework · Real-world AI breaches · Threat modelling for AI systems
Module 2
OWASP LLM Top 10 (2025 Edition)
Prompt injection · Insecure output handling · Training data poisoning · Model theft · Excessive agency · All 10 categories
Module 3
Prompt Injection & Jailbreaking
Direct injection · Indirect injection via RAG · Multi-turn manipulation · Jailbreak taxonomies · DAN & variants
Module 4
Automated Red Teaming: Garak & PyRIT
Garak probes & detectors · PyRIT orchestrators · Automated jailbreak generation · Coverage reporting
Module 5
Agentic AI Security
Tool abuse & privilege escalation · MCP security · Agent prompt injection · Memory poisoning · Multi-agent attack chains
Module 6
Adversarial ML: Evasion & Poisoning
Model extraction attacks · Membership inference · Data poisoning · Backdoor attacks · Adversarial examples
Module 7
AI Defences & Guardrails
Input validation · Output filtering · Lakera Guard · NeMo Guardrails · Prompt shields (Azure AI) · Red team vs blue team
Module 8
AI Security Assessment & Reporting
Scoping an AI pen test · Evidence collection · Risk rating for AI findings · Regulatory mapping (EU AI Act) · Report writing
Who this is for
- Security engineers and penetration testers upskilling into AI security
- AI engineers who want to understand how their systems can be attacked
- Red teamers adding LLM attack techniques to their toolkit
- Compliance officers managing AI risk under NACSA and EU AI Act
Tools & technologies
Prerequisites
- Cybersecurity fundamentals or penetration testing experience
- Familiarity with LLMs and prompt engineering concepts
- Basic Python for running tools