AI for the Defense

Here is the shape of our time together. Three ideas hold the whole session up.

First, ethics leads. Every technique I show you sits on top of your existing duties, so we start there and keep coming back to it.

Second, this is hands-on. You will spend more time working in a chatbot than listening to me. The scenarios are fictional, but they are built to feel like Monday morning.

Third, verification is the throughline. The goal is not to prompt faster. The goal is to catch the error before it reaches a judge. If you leave with one habit, I want it to be that one.

These are the threads we will pull on. You have all seen the headlines about lawyers sanctioned for citing cases that never existed, so we look hard at hallucinated case law and how it happens.

We look at confidentiality, because pasting a client fact into the wrong tool can be its own problem before a single word is filed.

Then we turn to the fixes. Structured prompting gives the model guardrails. Retrieval-augmented generation, which I will just call grounding, keeps the model tied to sources you trust instead of its own memory.

And underneath all of it sits your judgment. Keep that in mind as we go, and we will move to what you will actually be able to do by the end.

These are our objectives, and they are ethics-focused on purpose. Notice that not one of them is about becoming a faster typist.

You will be able to spot where AI creates risk, explain to a colleague or a bar committee why the responsibility stays with you, and recognize the moments where confidentiality and competence are in play.

Then the practical half. You will apply concrete guardrails, and you will use structured, source-grounded prompting to raise the accuracy of what you get back.

If you can do these five things on Monday, this hour was worth your time. Let me show you how we are going to spend it.

Here is the road map. We move in three segments. First, ethical foundations, where we name the duties and the failures. Second, the risk-reducing practices, which is where you do most of the hands-on work using a framework I call TCDLAi. Third, guardrails and takeaways, so you walk out with a checklist and clear rules.

The times on screen are targets, not promises. Each activity block is marked as core or optional. If we have the full ninety minutes, we run a second scenario. If we have sixty, we keep the core and still get you working. Either way, you leave having tried this yourself.

Next, the tools you will use, and the one rule that matters most before you type anything.

Let me get you set up. You do not need anything fancy. Pick one free chatbot. Claude, ChatGPT, Z.ai, and Mistral all have free tiers that will carry you through everything today.

If you did not set one up, do not worry. I am handing out a BoodleBox code so you can work alongside everyone else. And later we will glance at Claude for Legal as an example of tooling built specifically for our work, with attorney review built into it.

Now the rule in the box, and it is the most important sentence on this slide. Do not put confidential, privileged, or client-identifying information into a consumer chatbot. That is why every scenario I hand you today is invented. We practice safely first, then you carry the habit back to real matters.

One word about method, because it shapes everything we do. We are working problem-first, not tool-first.

Tool-first is the trap. You see a slick demo, you get excited, and you go looking for a reason to use it. That is how firms end up paying for software that solves nothing.

Problem-first flips it. We start with a defense task you already know is hard. We decide what a good result would look like. Then we point AI at it and, at the end, we ask the honest question, did this actually help. That last question is the whole game, and you will hear me ask it after every activity.

Before we start segment one, here is what you take home. Three things, all waiting on the resources page, so you do not have to write anything down.

The first is the TCDLAi Prompt Design Guide, a six-part structure built for defense work. The second is a one-page ethical guardrail checklist you run before anything leaves your desk. The third is a library of legal prompts, sorted by type and task, each one written to push you toward verification.

Bookmark that one address and the whole session is in your pocket. Now, before we open the ethics, let us warm up together and take our own measure.

Let us start by taking our own measure. Open the resources page and find the AI Fluency Self-Assessment. It is five quick questions, nothing is stored, and it lands you on one of four levels, from Disconnected to Strategic. Take two minutes and answer honestly. There are no wrong answers in this room.

Once you have your level, turn to your table. First, introduce yourselves. Name, the county you practice in, and one word for how AI feels in your work right now. Then compare your levels, and talk about what surprised you. Did the score match your gut.

Before we move on, each of you name one AI habit you want to build this year and say it out loud. Saying it to the table makes it real. Give them about five minutes, and listen for the themes, because they will come back all session. When you are ready, we begin with ethics.