AI Security in Production: Defending LLMs and ML Systems Against Prompt Injection, Adversarial Attacks, Model Poisoning, and Supply Chain Threats (Production AI Engineering Series) - Softcover

Inhaltsangabe

Protect Your AI Infrastructure Against Modern Adversarial Threats

As organizations rapidly deploy large language models (LLMs) and autonomous AI agents into production, a critical new attack surface has emerged. Traditional cybersecurity protocols—built for firewalls and APIs—are fundamentally blind to these specialized threats. AI Security in Production is your definitive, hands-on engineering guide to securing machine learning systems across the modern enterprise stack.

This comprehensive manual cuts through the theoretical hype to deliver deep technical analysis and practical defense frameworks. You will understand how adversaries exploit LLM vulnerabilities at a mechanistic level and, more importantly, how to build resilient mitigation architectures that preserve system performance.

• Defeating Prompt Injection: Neutralize direct jailbreaks and complex indirect injection attacks targeting Retrieval-Augmented Generation (RAG) pipelines.
• Securing the Model Supply Chain: Implement robust controls against training data poisoning, supply chain backdoors, and malicious pre-trained models from public repositories.
• Robustness Engineering: Mitigate adversarial examples, evasion tactics, and critical inference-time data leakage vulnerabilities.
• Agentic AI & MCP Security: Safely deploy autonomous agents and Model Context Protocol (MCP) servers interacting with production environments.
• DevSecOps & Compliance: Shift-left by integrating ML security into your CI/CD pipelines while aligning with NIST AI RMF, OWASP GenAI Top 10, and the EU AI Act.

Whether you are a security engineer, ML developer, or DevSecOps practitioner, this book equips you with the exact methodologies, red teaming frameworks, and production-ready open-source tools required to build resilient, trustworthy AI systems today.

Die Inhaltsangabe kann sich auf eine andere Ausgabe dieses Titels beziehen.