Trustworthy AI : Red Teaming, Runtime Assurance, and the Architecture of Secure Intelligence

  • 予約

Trustworthy AI : Red Teaming, Runtime Assurance, and the Architecture of Secure Intelligence

  • ただいまウェブストアではご注文を受け付けておりません。 ⇒古書を探す
  • 製本 Paperback:紙装版/ペーパーバック版/ページ数 170 p.
  • 言語 ENG
  • 商品コード 9781809204295

Full Description

Move beyond AI alignment claims. Build measurable runtime assurance with continuous red teaming, TIVM risk quantification, AI judges, observability, guardrails, and zero-trust architecture

Key Features

See how AI fails through prompt injection, jailbreaks, prompt leakage, and Crescendo escalation
Quantify AI risk with TIVM and turn assurance evidence into actionable risk signals
Build runtime assurance with continuous red teaming, AI judges, guardrails, and zero trust

Book DescriptionAI is moving from generating answers to making decisions and taking action, yet many organizations still rely on alignment claims, safety filters, and one-time tests to decide whether systems can be trusted. Trustworthy AI argues that assurance must be measurable, continuous, and enforced at runtime.

The book begins with how language-based systems fail under pressure: prompt injection, persona jailbreaks, system-prompt leakage, multi-turn Crescendo escalation, and bias. A red-team case study of a sovereign AI model shows why AI risk is not simply safe or failed, but often exists in a dangerous middle.

From there, the book moves from evidence to architecture. Sandeep Tulshiram Shilawat introduces the TIVM framework for quantifying AI risk and turning it into actionable assurance signals. He then presents an operating model built on continuous red teaming, AI observability, runtime guardrails, red-team agents, and independent AI judge systems. Final chapters connect these controls to evidence-based compliance, regulated and sovereign environments, zero-trust architecture, and Project NANDA.

For executives, security leaders, architects, engineers, regulators, and mission owners, this book provides a practical path from trusting AI by assertion to governing it through continuous evidence.What you will learn

Identify prompt injection, jailbreaks, and system-prompt leakage
Detect Crescendo escalation and bias under adversarial pressure
Quantify AI risk with TIVM and actionable assurance signals
Apply the NIST AI RMF in a simplified, practical way
Run continuous red teaming with autonomous red-team agents
Deploy AI observability, runtime guardrails, and AI judges
Build evidence-based compliance for regulated environments
Apply zero trust and sovereign-AI principles to AI systems

Who this book is forTrustworthy AI is for leaders and practitioners accountable for deploying, securing, governing, or overseeing AI. Five reading pathways guide executives and board members; CISOs, risk leaders, and security architects; AI engineers, platform owners, and agent architects; government, regulatory, and oversight teams; and defense, national-security, and sovereign-AI leaders to the chapters most relevant to their responsibilities.

Contents

Table of Contents

The Illusion of AI Alignment
AI as an Attack Surface
Case Study - Breaking a Sovereign AI Model
Prompt Injection - The Root of All Failures
Jailbreaks and Persona Attacks
System Prompt Leakage - Why Models Reveal Their Own Secrets
Multi-Turn Escalation Attacks - The Crescendo Problem
Bias, Sovereignty, and Political Risk
Why AI Risk Is Not Binary
NIST AI RMF Simplified
A Quantitative Model for AI Risk
Building an AI Risk Dashboard
Why One-Time Testing Fails
Continuous Red Teaming
AI Observability and Runtime Guardrails
Red Team Agent Systems
AI Judge Systems
Evidence-Based AI Compliance
AI in Government and Regulated Environments
Sovereign AI and National Security
Zero Trust for AI
Neocloud
Trust as a Competitive Advantage
Autonomous Governance

最近チェックした商品