Agentic AI Security Handbook: Design Patterns, Threat Models, and Defensive Controls for Autonomous LLM Agents (Paperback) (ISBN-13: 9798242151055)

Agentic AI Security Handbook is a practical, engineer-focused reference for securing autonomous LLM agents in real-world environments. It goes beyond theory to address the concrete risks introduced by agentic systems prompt injection inside planning loops, memory poisoning across sessions, tool abuse, data exfiltration, and uncontrolled execution.

This book is written for builders who need clear patterns, threat models, and defensive controls not abstract discussions or vendor marketing.

• How agentic architectures expand the attack surface beyond prompts

• Threat modeling techniques tailored specifically for autonomous agents

• Prompt injection risks in multi-step reasoning and tool execution

• Memory poisoning, persistence attacks, and long-term state manipulation

• Tool and API abuse, permission escalation, and unsafe tool chaining

• Data and RAG security risks in agent-driven retrieval workflows

• Runtime guardrails, policy enforcement, approval gates, and kill switches

• Red-teaming, evaluation, and continuous security testing for agents

• An enterprise-grade reference architecture for secure agentic systems

Most AI security books focus on LLMs as static components. This handbook treats AI as what it has become: an autonomous system capable of action.

Every chapter is designed to be:

• Practical and implementation-oriented

• Structured for quick reference and reuse

• Focused on why controls exist, not just what they are

Each section ends with checklists, failure modes, and actionable guidance you can apply immediately.

• AI and ML engineers building agentic systems

• Security engineers responsible for AI risk

• Platform and infrastructure architects

• Technical leaders deploying autonomous AI in production

If you are designing, deploying, or securing AI agents that can reason, act, and interact with real systems, this handbook belongs on your desk.

Build smarter agents. Harden them properly. Keep control.