Agentic AI Defense Architecture: Secure Autonomous LLM Agents with Threat Models, Safe Prompting, Memory Controls & Continuous Hardening

By Morgan Liang

Autonomous LLM agents represent the most disruptive shift in modern software engineering. Unlike traditional applications, agentic systems can reason, plan, call external tools, write to memory, and run multi-step workflows without direct human oversight. This new level of autonomy introduces massive opportunity and equally significant risk.
Defensive Agentic AI Engineering is a complete, practitioner-oriented guide to building and securing the next generation of intelligent systems. Designed for AI engineers, security architects, builders of LangGraph-style agents, workflow automation developers, and anyone deploying autonomous reasoning systems, this book teaches you how to defend agentic architectures before vulnerabilities reach production.
Inside this handbook, you'll learn how to design autonomous agents that are powerful, reliable, and hardened against adversarial manipulation. From prompt-layer protections to memory-integrity controls and tool-invocation guardrails, you'll discover proven defensive patterns that keep agents aligned with user intent - even when exposed to unpredictable real-world inputs.
What You Will Learn
Agentic threat modeling using AI-tailored security frameworks and analysis patterns
Secure prompt engineering using schemas, structured outputs, role conditioning, and drift-prevention controls
Tool-use security including action gating, sandbox isolation, execution boundaries, and least-privilege design
Memory safety techniques such as integrity scoring, write filters, recall validation, and poisoning resistance
Closed-loop agent defense with safety critics, self-evaluation hooks, policy checks, and reasoning-loop enforcement
Adversarial testing and red-teaming for autonomous AI, covering multimodal inputs, jailbreak attempts, and recursive exploitation
Compliance alignment with NIST AI RMF, OWASP GenAI Top 10, and global safety requirements for enterprise deployment
.
Who This Book Is ForAI developers and engineersSecurity researchers and DevSecOps teamsBuilders of agentic systems and automation workflowsTechnical founders integrating LLM agents into productsResponsible AI practitioners ensuring safe deploymentThe future of AI is autonomous and its safety depends on engineers who understand how to defend it.
Start building trustworthy, resilient agentic systems today.

• Format: Paperback
• Language: English
• ISBN: B0G4QWTZGM
• ISBN13: 9798276943749
• Release Date: December 2025
• Publisher: Independently Published
• Length: 104 Pages
• Weight: 0.43 lbs.
• Dimensions: 0.2" x 7.0" x 10.0"