What Can You Achieve with This Training?

Learn to systematically assess Large Language Models (LLMs) and generative AI applications for security vulnerabilities.
Gain hands-on experience investigating prompt injection, jailbreak, and data exfiltration scenarios in controlled laboratory environments.
Analyse emerging security threats involving AI agents, RAG, and MCP-based systems, including risks introduced by untrusted data.
Develop practical knowledge of automated AI security assessments using PyRIT, Inspect AI, and AutoGen.
Explore Responsible AI, threat modelling, and mitigation strategies to evaluate the security and responsible use of AI systems.
Reinforce your skills through hands-on exercises and competitive CTF-style challenges using the Tinycode Cyber Range platform.

The AI Red Teaming Training is a two-day, hands-on programme designed for security professionals who want to systematically evaluate the security of modern artificial intelligence systems from an adversarial perspective.

The course introduces practical approaches to AI Red Teaming, enabling participants to investigate vulnerabilities in large language models and AI-powered applications, assess different attack methods, and examine strategies for reducing identified risks.

Key topics include adversarial machine learning, Responsible AI violations, and emerging threats to autonomous AI agents. The programme begins with an overview of classical machine learning, deep learning, and Transformer architectures before exploring security considerations in modern AI components such as Retrieval-Augmented Generation (RAG) and the Model Context Protocol (MCP).

Attack-focused modules examine direct and indirect prompt injection, jailbreak techniques, data exfiltration, memory poisoning, and tool hijacking. Participants also explore advanced approaches including Crescendo, Skeleton Key, GCG, PAIR, and TAP.

The training introduces open-source tools such as PyRIT, Inspect AI, and AutoGen for automated AI security assessments. It also examines security risks in multimodal AI systems that process text, images, audio, and video, including misinformation, bias, and harmful content scenarios.

Throughout the programme, participants use the Tinycode Cyber Range to investigate realistic security scenarios in a controlled environment. CTF-style challenges, scoring, and leaderboards provide opportunities to assess practical understanding.

The final modules focus on AI security controls, defensive techniques, governance, and risk mitigation. Participants develop a broader understanding of how to identify AI vulnerabilities and evaluate approaches for reducing their impact.

For corporate groups, we can organize this training as an on-site private group.

Prerequisites

No previous experience in Machine Learning or Data Science is required.

Participants should have:

A solid understanding of general cybersecurity principles
Basic scripting knowledge using Python or Bash

Who Should Attend

This course is suitable for professionals who want to assess and improve the security of AI-powered systems.

It is particularly relevant to:

Cybersecurity professionals
Security consultants
Red Team operators
Teams looking to add AI vulnerability assessment to their existing security capabilities
Security professionals seeking to understand the logic, data flows, and unique risks of LLM-integrated systems

What You Will Learn

By the end of this course, participants will be able to:

Systematically assess LLM applications for prompt injection, data extraction, and jailbreak vulnerabilities.
Explain and test direct and indirect prompt injection techniques in controlled environments.
Evaluate advanced attack methods, including Crescendo, Skeleton Key, GCG, PAIR, and TAP.
Identify security risks in RAG, MCP, and AI agent data flows.
Analyse agentic AI threats such as memory poisoning and tool hijacking.
Use PyRIT, Inspect AI, and AutoGen to support automated AI security assessments, benchmarking, and agent development.
Compare open-box and closed-box approaches and explain the differences between semantic and token-level attacks.
Assess risks related to bias, misinformation, and harmful content in multimodal AI systems.
Identify Responsible AI violations involving bias, privacy, misinformation, and harmful content.
Contribute to planning AI Red Team activities using the Map, Measure, Mitigate framework.
Explain defensive approaches such as Defensive Prompting, Spotlighting, and NVIDIA NeMo Guardrails, including their limitations.
Gain practical experience conducting security assessments within the Tinycode Cyber Range.
Relate identified AI vulnerabilities to appropriate risk mitigation and defensive strategies.

Training Outline

1. Welcome and Lab Environment

Introducing the training schedule, laboratory exercises, and platform
Accessing the Tinycode Cyber Range
Running controlled scenarios involving direct prompt injection and data exfiltration

2. Introduction to Machine Learning and AI

Reviewing classical machine learning, deep learning, and Transformer architectures
Understanding the data-versus-code problem in Large Language Models
Exploring the modern AI technology stack:
- Retrieval-Augmented Generation (RAG)
- Model Context Protocol (MCP)
- Common failure modes and security weaknesses

3. Introduction to Attacking Generative AI

Exploring direct and indirect prompt injection techniques
Examining jailbreak approaches, including Crescendo and Skeleton Key
Understanding encoding tricks and persuasion-based attack patterns
Hands-on laboratory exercises involving:
- System prompt extraction scenarios
- Risks associated with generating prohibited or harmful content
- Chaining multiple attack techniques

4. AI Agents and Agentic Systems

Examining how agentic AI changes existing threat models
Exploring memory poisoning, tool hijacking, and OpenClaw-related failures
Completing laboratory exercises involving AI agents processing untrusted data

5. Open-Source Tools for AI Red Teaming

Introducing automated AI security testing tools:
- PyRIT
- Inspect AI
- AutoGen
Building and running automated attack workflows
Exploring applications in security assessment, benchmarking, and agent development

6. Advanced Attack Techniques

Examining algorithmic jailbreaking approaches
Comparing open-box and closed-box attack methods
Exploring advanced techniques:
- Greedy Coordinate Gradient (GCG)
- Prompt Automatic Iterative Refinement (PAIR)
- Tree of Attacks with Pruning (TAP)
Evaluating trade-offs between semantic and token-level attacks
Implementing PAIR from scratch in a practical exercise

7. Multimodal Models

Exploring the evolution of AI models that process text, images, audio, and video
Examining security challenges and compounded biases arising from multiple input types
Completing laboratory exercises involving multimodal misinformation and propaganda scenarios

8. Responsible AI and the AIRT Process

Understanding the principles of fairness, transparency, and accountability
Developing an AI Red Team programme using the Map, Measure, Mitigate framework
Discussing the human impact of AI development

9. Mitigations

Reviewing technical and procedural security controls for AI systems
Exploring defensive approaches:
- Defensive Prompting
- Spotlighting
- NVIDIA NeMo Guardrails
Examining the limitations of technical controls against prompt injection

10. AI, Science, and Open Discussion

Evaluating high-stakes AI applications in scientific research and national security
Reviewing a case study involving Anthropic's Claude Mythos and Project Glasswing
Discussing emerging challenges and developments in AI security

Why Choose Us

Leading UK-based global training provider since 1995.

Experience AI Red Teaming through Bilginç IT Academy's live and interactive virtual classroom environment. Join a Public course as an individual delegate or arrange a dedicated Private / In-house online training program exclusively for your organization.

Delivery Method: Online Instructor-Led
Participation Model: Public / Private (In-house)
Live and Interactive Training: Connect with your instructor in real time and actively participate through discussions, Q&A sessions, practical exercises, and group activities.
Flexible Participation: Join the training from your home, office, or any location with a suitable internet connection.
Expert Trainer Network: Learn from experienced trainers with strong industry backgrounds and practical field expertise.
Over 30 Years of Training Expertise: Benefit from Bilginç IT Academy's professional training experience since 1995.
Worldwide Access: Join our live virtual classrooms from anywhere in the world or arrange a dedicated online training program for your organization.

Experience AI Red Teaming in a professional face-to-face classroom environment. Classroom Based training can be delivered as a Public course open to individual delegates or as a dedicated Private / In-house class exclusively for your organization.

Delivery Method: Classroom Based
Participation Model: Public / Private (In-house)
Face-to-Face Learning: Interact directly with your instructor and fellow delegates in an engaging classroom environment.
Experienced Trainers: Learn from specialists with extensive industry experience and practical real-world knowledge.
Professional Training Environment: Attend training in comfortable, well-equipped classrooms designed to support effective learning.
Practical Learning: Depending on the course, reinforce your knowledge through hands-on exercises, scenarios, case studies, and instructor-led activities.

Arrange AI Red Teaming as a dedicated Onsite training program for your organization. Bilginç IT Academy trainers can deliver the training at your office or another location of your choice, with the program planned around your team's requirements and business objectives.

Delivery Method: Onsite
Participation Model: Private (In-house)
Training at Your Preferred Location: Organize the training at your company's office or another location selected by your organization.
Tailored Course Content: Adapt the training program to your projects, team structure, existing skill levels, and specific business requirements.
Team-Focused Learning: Develop your team around a shared knowledge base while strengthening internal collaboration and knowledge transfer.
Flexible Scheduling: Plan the training dates, location, and program according to your organization's operational requirements.
Worldwide Onsite Delivery: Bilginç IT Academy trainers can travel internationally to deliver dedicated training programs at your preferred location.

AI Red Teaming Training