Red Teaming AI. A Field Manual for Attacking Intelligent Systems

AI is no longer a futuristic concept—it’s embedded in critical systems shaping finance, healthcare, infrastructure, and national security. But with this power comes unprecedented risk. Red Teaming AI arms you with the mindset, methodology, and tools to proactively test and secure intelligent systems before real adversaries exploit them.

Written for security professionals, researchers, and AI practitioners, this field manual goes beyond theory. You’ll learn how to map the new AI attack surface, anticipate adversarial moves, and simulate real-world threats to uncover hidden vulnerabilities.

You’ll Learn How To:

Think in graphs, not checklists: trace attack paths through interconnected AI components, data pipelines, and human interactions

Poison the well: explore how adversaries corrupt training data to implant backdoors and erode model integrity

Fool the oracle: craft evasion attacks that manipulate AI perception at decision time

Hijack conversations: execute prompt injection attacks that turn Large Language Models into insider threats

Steal the brain: probe for model extraction and privacy attacks that compromise valuable IP

Conduct full-spectrum campaigns: use the STRATEGEMS framework and the AI Kill Graph to plan, execute, and report professional-grade red team engagements

Traditional security methods can’t keep up with adversarial AI. From manipulated financial agents to compromised autonomous vehicles, real-world failures have already caused billions in losses and threatened lives. Red Teaming AI equips you to meet this challenge with practical techniques grounded in real attack scenarios and cutting-edge research.

Table of contents

Introduction

Part I: The Adversarial Playbook: Mindset & Methodology

Chapter 1: The New Attack Surface: Thinking in Graphs

Chapter 2: The Engagement: An AI Red Teamer's Methodology

Part II: The AI Kill Graph: Core Attack Techniques

Chapter 3: Reconnaissance: Mapping the AI Terrain

Chapter 4: Poisoning the Well: Corrupting AI Data

Chapter 5: Fooling the Oracle: Evasive Attacks at Inference

Chapter 6: Hijacking the Conversation: LLM Prompt Injection

Chapter 7: Seizing Control: Agentic System Exploitation

Chapter 8: Stealing the Brain: Model Extraction and Privacy Attacks

Part III: The Campaign: Execution & Impact

Chapter 9: Graphs of Pain: Advanced Attack Sequences

Chapter 10: The Endgame: Reporting for Maximum Impact

Chapter 11: The Next Frontier: The Future of AI Red Teaming

References