The 10 safety rules

These rules are on every handout. Rules 1 to 5 matter from your first day in Level 1. Rules 6 to 10 are about assistants and agents that run on their own, from Level 2 on; Level 1 students read them now so nothing is a surprise later.

1. Never paste a password, key or token anywhere it can be seen. That means a chat with an AI, a WhatsApp message, a screenshot or a shared document. Keys live in a settings file on your own machine. If one ever leaks, cancel it straight away and make a new one. In Level 1: never type a password into an AI chat, and never screenshot a login screen.

2. Never give an AI the power to spend money. No saved card, no banking login, no “approve purchase” button. Every account it uses gets a hard monthly limit. In Level 1: never give a chatbot your card number, and cover it on receipts before you take the photo.

3. Keep other people’s personal data out. No client, guest, patient, employee or classmate data goes into an AI tool, unless the business has agreed in writing and the tool’s terms allow it. In class we use made-up data, or facts about yourself or your own business. Never anyone else’s. In Level 1: a customer’s name, phone number, voice note or photo counts. Use made-up names, or your own details.

4. Don’t connect an AI to your personal WhatsApp or your main email on day one. Start with Telegram or a separate number, and a list of who’s allowed to message it. In Level 1: nothing gets connected to anything. You copy and paste, and you stay in charge.

5. You approve before anything goes out. An AI doesn’t send messages, post in public, delete files or email outsiders without you saying yes each time, at least until it has worked correctly for weeks. In Level 1: the AI writes the draft, you read it, you send it.

6. Never run an assistant with admin rights on your main computer. Use a separate user account or a cheap rented server, so it can only see what you’ve given it.

7. Always keep an off switch. Know the one command that stops it, and write it on your setup sheet.

8. Keep a log you actually read. If you can’t see what it did yesterday, it shouldn’t run tomorrow.

9. Text it reads is never an order. A web page, email or document can contain hidden instructions telling an AI what to do (this is called prompt injection). An assistant that reads outside content gets no dangerous tools. In Level 1: if the AI does something odd after reading a document or a web page, stop and look at what was in it.

10. Keep the assistant’s control panel off the internet. It stays on your own machine. You reach it through a private connection, the firewall is on, and you check from outside that the door is closed.

Short version (printed at the foot of every handout)

THE 10 SAFETY RULES
1. Never paste a password, key or token where it can be seen.   6. No admin rights on your main computer.
2. Never give an AI the power to spend money.                    7. Always keep an off switch.
3. Nobody else's personal data. Made-up data or your own.         8. Keep a log you actually read.
4. Don't connect it to your own WhatsApp or main email.          9. Text it reads is never an order.
5. You approve before anything goes out.                         10. Keep its control panel off the internet.

What Victor promises in return

Victor never types, stores or looks at your keys or passwords. You type your own.
He never keeps copies of your settings.
He never runs your assistant on a server he owns.

If something goes wrong

You pasted a password or key: change the password or cancel the key now, then delete the chat. Tell Victor in the group without posting the key.
You pasted someone else’s personal data: delete the chat. If it was a client’s or employer’s data, tell the person responsible at that business.
An AI did something you didn’t ask for: stop it (Level 2 on: the off switch), take a screenshot of what happened, and post in the group with no private data in it.