menu
{ "item_title" : "Threat Modeling for Generative AI", "item_author" : [" Mavik Orswell "], "item_description" : "The attacks on generative AI systems are real - PoisonedRAG achieved a ninety-percent attack success rate in production knowledge bases. EchoLeak exfiltrated SharePoint data through Microsoft Copilot with no user interaction. The Replit coding agent deleted a production database despite explicit instructions not to change anything. Most teams applying traditional security reviews to these systems are looking at the wrong surface. This book gives security engineers, AI engineers, and DevSecOps practitioners a structured, practitioner-grade methodology for finding threats in the three deployment patterns that define enterprise GenAI risk: RAG pipelines, fine-tuned models, and agentic workflows. Using the GATE framework (Generative AI Threat Enumeration), MITRE ATLAS v5.4.0, MAESTRO, and the OWASP LLM and Agentic Top 10, you will build threat models that reflect how real attacks work - not how they look in whitepapers. - Map attack surfaces across the full generative AI stack: training data, embeddings, retrieval index, inference, and agent action layers- Apply the MAESTRO 7-layer framework to identify trust boundaries and threat chains in agentic AI systems- Design RAG pipeline security controls from ingestion filtering through retrieval authorization to output validation- Evaluate fine-tuning datasets and workflows for data poisoning, backdoor injection, and LoRA adapter vulnerabilities- Construct OWASP LLM Top 10 threat models for production LLM applications and RAG systems- Detect and defend against prompt injection - direct and indirect - using layered architectural controls- Apply MITRE ATLAS tactics and techniques to build attacker-profile-driven threat scenarios for GenAI systems- Implement runtime monitoring, guardrails, and observability for agentic workflows with tool access and persistent memory- Conduct structured red-team assessments of RAG pipelines, fine-tuned models, and multi-agent architectures- Build a repeatable threat modeling process that integrates into AI and ML development pipelines The GATE methodology sequences STRIDE, MITRE ATLAS, OWASP, and LINDDUN in layer order across the five-layer GenAI stack, producing one deduplicated threat register in a single two-to-four-hour session. Every chapter includes verified real-world case studies, runnable code, and specific decision filters for your architecture. The EU AI Act compliance mapping in Chapter 11 shows how GATE outputs satisfy most Annex IV documentation requirements without additional work. For security engineers, AI/ML engineers, and DevSecOps practitioners who build or secure production GenAI systems and need a systematic, architecture-specific method for finding threats before the breach report does. Stop waiting for the incident. Start with the threat model.", "item_img_path" : "https://covers3.booksamillion.com/covers/bam/9/79/816/748/9798167486294_b.jpg", "price_data" : { "retail_price" : "29.99", "online_price" : "29.99", "our_price" : "29.99", "club_price" : "29.99", "savings_pct" : "0", "savings_amt" : "0.00", "club_savings_pct" : "0", "club_savings_amt" : "0.00", "discount_pct" : "10", "store_price" : "" } }
Threat Modeling for Generative AI|Mavik Orswell

Threat Modeling for Generative AI : Designing Security into RAG, Fine-Tuning, and Agentic Workflows

local_shippingShip to Me
In Stock.
FREE Shipping for Club Members help

Overview

The attacks on generative AI systems are real - PoisonedRAG achieved a ninety-percent attack success rate in production knowledge bases. EchoLeak exfiltrated SharePoint data through Microsoft Copilot with no user interaction. The Replit coding agent deleted a production database despite explicit instructions not to change anything. Most teams applying traditional security reviews to these systems are looking at the wrong surface.

This book gives security engineers, AI engineers, and DevSecOps practitioners a structured, practitioner-grade methodology for finding threats in the three deployment patterns that define enterprise GenAI risk: RAG pipelines, fine-tuned models, and agentic workflows. Using the GATE framework (Generative AI Threat Enumeration), MITRE ATLAS v5.4.0, MAESTRO, and the OWASP LLM and Agentic Top 10, you will build threat models that reflect how real attacks work - not how they look in whitepapers.

- Map attack surfaces across the full generative AI stack: training data, embeddings, retrieval index, inference, and agent action layers
- Apply the MAESTRO 7-layer framework to identify trust boundaries and threat chains in agentic AI systems
- Design RAG pipeline security controls from ingestion filtering through retrieval authorization to output validation
- Evaluate fine-tuning datasets and workflows for data poisoning, backdoor injection, and LoRA adapter vulnerabilities
- Construct OWASP LLM Top 10 threat models for production LLM applications and RAG systems
- Detect and defend against prompt injection - direct and indirect - using layered architectural controls
- Apply MITRE ATLAS tactics and techniques to build attacker-profile-driven threat scenarios for GenAI systems
- Implement runtime monitoring, guardrails, and observability for agentic workflows with tool access and persistent memory
- Conduct structured red-team assessments of RAG pipelines, fine-tuned models, and multi-agent architectures
- Build a repeatable threat modeling process that integrates into AI and ML development pipelines

The GATE methodology sequences STRIDE, MITRE ATLAS, OWASP, and LINDDUN in layer order across the five-layer GenAI stack, producing one deduplicated threat register in a single two-to-four-hour session. Every chapter includes verified real-world case studies, runnable code, and specific decision filters for your architecture. The EU AI Act compliance mapping in Chapter 11 shows how GATE outputs satisfy most Annex IV documentation requirements without additional work.

For security engineers, AI/ML engineers, and DevSecOps practitioners who build or secure production GenAI systems and need a systematic, architecture-specific method for finding threats before the breach report does.

Stop waiting for the incident. Start with the threat model.

This item is Non-Returnable

Details

  • ISBN-13: 9798167486294
  • ISBN-10: 9798167486294
  • Publisher: Independently Published
  • Publish Date: November 2026
  • Dimensions: 11 x 8.5 x 0.55 inches
  • Shipping Weight: 1.34 pounds
  • Page Count: 260

Related Categories

You May Also Like...

    1

BAM Customer Reviews