AI Security
Attacks, Defences and Secure Design for AI Systems
- Pages
- 82
- Chapters
- 66
- Level
- Intermediate–Advanced
- Format
- E-book
About this book
AI systems add new attack surfaces to familiar software: the data they learn from, the text they read and the actions they can take. This book explains the threats and the defences that work.
You will learn threat modelling for AI, prompt injection and jailbreaks, data poisoning and backdoors, adversarial examples, model and data extraction, privacy attacks, supply-chain risks in models and datasets, and the special risks of agents with tools. Defensive chapters cover secure architecture, least privilege, input and output controls, monitoring, red teaming and incident response.
The focus is defensive and practical: understanding attacks so you can design systems that resist them.
What you will learn
- Threat modelling for AI systems
- Prompt injection, jailbreaks and their limits as defences
- Data poisoning, backdoors and adversarial examples
- Model extraction, data leakage and privacy attacks
- Supply-chain security for models and datasets
- Secure architecture, monitoring, red teaming and incident response
Table of contents
- IntroductionIntroduction
- Chapter 1Why AI Security Is Different
- Chapter 2Security Principles That Still Apply
- Chapter 3Threat Modelling for AI Systems
- Chapter 4The AI Attack Surface
- Chapter 5Attacker Goals
- Chapter 6Risk Depends on Capability
- Chapter 7Security Requirements Up Front
- Chapter 8Shared Responsibility With Providers
- Chapter 9Prompt Injection
- Chapter 10Indirect Prompt Injection
- Chapter 11Why Prompt Injection Is Hard to Solve
- Chapter 12Defending Against Prompt Injection
- Chapter 13The Lethal Combination
- Chapter 14Jailbreaks
- Chapter 15System Prompt Leakage
- Chapter 16Sensitive Data Disclosure
- Chapter 17Securing Retrieval Systems
- Chapter 18Agents and Tool Use
- Chapter 19Insecure Output Handling
- Chapter 20Denial of Service and Cost Abuse
- Chapter 21Data Poisoning
- Chapter 22Backdoors
- Chapter 23Defending Training Pipelines
- Chapter 24Adversarial Examples
- Chapter 25Evasion of AI-Based Security Tools
- Chapter 26Model Extraction
- Chapter 27Training Data Extraction
- Chapter 28Membership Inference
- Chapter 29Model Inversion
- Chapter 30Differential Privacy
- Chapter 31Federated and Confidential Computing
- Chapter 32Watermarking and Fingerprinting Models
- Chapter 33The AI Supply Chain
- Chapter 34Malicious Model Files
- Chapter 35Dependency Management
- Chapter 36Model and Dataset Provenance
- Chapter 37Plugins and Integrations
- Chapter 38Infrastructure Security
- Chapter 39Secrets Management
- Chapter 40Identity and Access for AI Systems
- Chapter 41Multi-Tenant Isolation
- Chapter 42Secure Architecture Patterns
- Chapter 43Input Controls
- Chapter 44Output Controls
- Chapter 45Human Oversight
- Chapter 46Logging and Monitoring
- Chapter 47Red Teaming
- Chapter 48Security Evaluation Suites
- Chapter 49Incident Response for AI
- Chapter 50Kill Switches
- Chapter 51Governance and Policy
- Chapter 52Standards and Frameworks
- Chapter 53Training People
- Chapter 54AI-Enabled Attacks
- Chapter 55Using AI for Defence
- Chapter 56A Threat Model, Worked Through
- Chapter 57Secure Prompt Design
- Chapter 58Testing Agents in Sandboxes
- Chapter 59Securing Fine-Tuning and Training
- Chapter 60Images, Audio and Documents as Inputs
- Chapter 61Protecting AI Logs
- Chapter 62Measuring Security Posture
- Chapter 63Case Study: The Email Assistant
- Chapter 64Case Study: The Leaky Knowledge Base
- Chapter 65Case Study: The Poisoned Feedback Loop
- Chapter 66A Security Checklist
- GlossaryGlossary
Who this book is for
Security engineers, developers and architects who build or protect AI systems.
Details
| Format | PDF + web edition |
|---|---|
| Length | 82 pages (approx.) |
| Chapters | 66 |
| Level | Intermediate to Advanced |
| Language | English |
| Topic | Ethics & Governance |
| Author | Latent Library Editorial |
Keep reading