ntworld.ink
Blue Team Manual

The Shadow AI Blue Team Manual

The computer science a defender needs, taught through one question: why can an employer see that a staff member reached a chatbot, but not what they typed into it?

Written manual Five parts For early-career defenders

Australian cybersecurity courses usually teach networking, identity, operating systems, cloud and governance as separate units, often through overseas textbooks and case studies. Students come out able to recite the OSI layers but unable to answer a question a manager might ask in a meeting. This manual teaches the same fundamentals through one current problem: staff using consumer AI tools such as ChatGPT, Claude and Gemini at work, and what an organisation can and cannot see, log or block when they do.

The problem spans almost every layer of the stack at once: how names resolve to addresses, how traffic is encrypted, how a browser isolates a tab, how a work account signs in to a third-party app, how a device reports that it is managed, and how the law constrains monitoring in an Australian workplace. Understanding the shadow AI control problem covers a large part of modern defensive security.

The manual is institution-agnostic. Microsoft 365 appears often, because it is the dominant enterprise stack in Australian higher education and government, but the grounding generalises, and where a non-Microsoft tool illustrates a concept more clearly it is used instead. This is not a product tour, and it is not a substitute for legal advice; the governance part teaches the legal landscape so a defender can hold a competent conversation with a privacy officer, not replace one.

Who it is for

Entry-level trainees and career-transitioning students: readers who will work through a chapter but are not yet fluent in the terminology. Each chapter assumes no prior exposure and defines its terms as it goes. Read the parts in order, or start with the one that matches your current gap.

How the manual is organised

There are five parts. Each chapter explains one concept, then shows how it determines what a control can and cannot do. A single glossary covers the whole manual; a term is set in bold on first use in a chapter and defined there.

Start here

The manual is complete: thirty-three chapters across five parts, from the OSI and TCP/IP models through to incident response when shadow AI goes wrong. The first chapter sets out the method used throughout. The glossary defines every term in plain language.

Last updated 9 August 2026