AI security training

AI can help us notice, organize, draft, and learn. This module is not about whether to use it. It is about what information you put into it. Before company information goes into an AI tool, give it a color.

Use AI to make the work better. Never use it to hand away responsibility.

Three colors

Three kinds of AI tool

Watch what you add along the way. An AI output inherits the highest classification of any input that produced it, so a Green task turns Red the moment you paste Red information into it.

AI can help you research, draft, and narrow a search. It can also be wrong in a way that sounds exactly like being right. This module is about two habits: asking for exactly what you need, and checking what comes back before you rely on it.

Use AI to make the work better. Never use it to hand away responsibility.

A good prompt has four parts

1. 1 Goal What you are trying to get done, and what the result is for. …for the job of [what the part must do]
2. 2 Exact specifications The model, version, dates, or numbers that make it your situation, not the average one. a 2023 [manufacturer] [model], [specific version/configuration]
3. 3 Constraints What it must not assume, guess, or promise, and what to do when it is unsure. Do not assume stock or pricing. Mark any uncertain information clearly.
4. 4 Desired output The shape you need back, including sources you can check yourself. List current manufacturer part numbers… Provide official sources

Then verify before you rely on it

Check anything that changes over time against the source of truth: the manufacturer, the distributor, the official policy, the system of record. Not against the AI.

• Compatibility
• Stock
• Price
• Dates and deadlines
• Versions
• Regulations
• Quotes and sources
• Calculations

AI can narrow the search. It is not the inventory system, the policy, or the final authority. Assume it may be confidently wrong until you have checked. A confident tone is a writing style, not evidence.

This one is for managers, department heads, and executives. A chatbot answers a question; an agent takes action. It reads an email, writes a record, sends a message. You decide which agents get built, what they can reach, and what happens when one goes wrong.

AI is a tool. Accountability is a choice. Leadership is making that choice visible.

The autonomy ladder

Most agents should live at Stage 1 or 2. Stage 3 is earned. Stage 4 should be rare.

Three kinds of question

Grant the minimum permissions necessary, and revoke them when the work is done. Test in a sandbox before anything reaches production. Build the audit trail before you need it.