Where do you begin?
Who This Is For
This guide is for anyone who uses AI — or wants to — but doesn’t yet have a reliable way to judge when its output is safe, useful, wrong, incomplete, or risky. AI is genuinely capable, and this manual is not here to talk you out of using it; it’s here to give you the judgment to use it well.
That includes hobbyists and new users, students, solo builders and creators, employees working on their own, and managers or small-business owners responsible for how others use AI.
Why This Guide Exists
I built this guide, and I’ll be upfront: I’m not a computer scientist, a lawyer, or an AI researcher. I started using AI to build something real — an AI-assisted trading system — with no prior experience, and I assumed that if the output looked right, it probably was. It didn’t, and it wasn’t. The work was fluent, confident, and wrong in ways I only caught after building the checks I should have had from the start.
That’s the whole reason this exists. And because I’m not an authority, I don’t ask you to take my word for any of it — every risk here is tied to a primary source you can verify yourself: NIST, OWASP, published research, and court records. This is the field guide I wish I’d had before I started.
Before You Continue
This page is a metacognitive assessment — it asks what you currently know and how you currently work with AI. The Risk Ladder, later in this manual, is the operational counterpart: a framework for evaluating individual AI outputs at the point of use.
Take a moment. There are no right answers here and nothing to submit. These questions are for you.
When an AI tool gives you an answer, do you have a general sense of how it arrived at that answer — or does it feel like it comes from a black box?
Could you explain to someone else, in plain terms, what an AI language model actually does when it responds to a question?
Have you ever been surprised — or misled — by something an AI tool got wrong? What did that look like?
When you read AI-generated output, do you find yourself looking for signs that something might be off — or do you tend to read it the same way you would read any other source?
Before acting on AI output, what do you currently do to check it? Is that process consistent, or does it depend on how confident the AI sounds?
If an AI tool gave you a well-written, confident answer, would you be more likely to trust it without checking — simply because it reads as authoritative?
If the AI output you are relying on turned out to be wrong, what would the consequence be — for you, for someone else, or for a decision already made?
Do you currently have any process for what to do when AI output is wrong or incomplete — or is this something you have not yet needed to think about?