AI Security in Practice

Key benefits

• Map AI assets, data flows, and trust boundaries across a complete application.
• Recognize data poisoning, adversarial attacks, prompt injection, and model extraction.
• Protect sensitive information in models, retrieval systems, and connected services.
• Assess dependencies and supplier risks throughout the AI supply chain.
• Design production controls for authorization, monitoring, incident response, and recovery.
• Connect technical decisions to governance, accountable ownership, and responsible AI.

Description

An AI assistant can retrieve internal records, interpret a document, and propose an account change. Each connection raises a practical security question: what is the system allowed to know and do?

AI Security in Practice explains how to protect the data, models, applications, and operational processes behind AI systems. It gives architects and security teams a clear way to examine exposure, choose controls, and ask for evidence before a capability reaches production.

The book begins with the foundations of AI security, then explores data poisoning, adversarial attacks, prompt injection, model extraction, privacy risks, and the AI supply chain. Later chapters bring these subjects together through production architecture, incident response, governance, and responsible AI, including the European regulatory context.

Throughout the book, the fictional Northbridge Services organization introduces a predictive transaction model, an internal knowledge assistant, and a service desk agent. These connected examples show how a broad service account, an untrusted document, or an insufficiently restricted tool can create consequences beyond the model itself.

A central principle runs through the chapters: identity services, application permissions, and independent policy controls must enforce authority. A model may propose an answer or action; the surrounding system must decide what is permitted.

Diagrams and conceptual illustrations make the relationships easier to follow. Chapter review exercises, practical worksheets, a glossary, and references help readers turn the material into questions for their own designs and operating processes. You can follow the book without experience training a large model; familiarity with applications, data flows, and connected services is enough to get started.

Who this book is for

Written for solution architects, security professionals, developers, data teams, and technical leaders who need to evaluate and operate AI in organizations.