Overview
AI systems have moved beyond generating text into taking action. They're in production. They query internal data, make API calls, and interact with other production systems, often with more access than most humans get. AI systems aren't deterministic; they reason, adapt, and operate on untrusted input in ways that traditional security models simply weren't designed for. This creates new vulnerabilities and shifts the entire control surface. This book is about that shift.
In Securing AI Systems, Dan Borges and David Campbell show you how to rethink security for AI systems built on retrieval pipelines, persistent memories, and agents that take action. Drawing from real-world adversarial testing and production deployments, they focus on how these systems actually fail: prompt injection that turns inputs into instructions, poisoned retrieval that corrupts decisions at runtime, and agents that quietly accumulate more authority than intended. Rather than relying on the model to "do the right thing," you'll learn how to design systems that constrain what AI systems are allowed to do, enforce least privilege at the capability level, and build architecture that can observe, interrupt, and contain failures when they happen.
- Understand why AI systems change traditional trust boundaries
- Learn how probabilistic models alter your threat model
- Identify where modern AI systems break in production
- Design systems with least privilege and controlled blast radii
- Apply zero trust boundaries and secure deployment patterns
Become an O’Reilly member and get unlimited access to this title plus top books and audiobooks from O’Reilly and nearly 200 top publishers, thousands of courses curated by job role, 150+ live events each month,
and much more.
Read now
Unlock full access