Skip to content

Unit 6: Keep it draft-only · Lesson 2 of 3 · about 7 min

Run a security check that changes nothing

Your AI agent can use a shared computer and your accounts. You want a list of risks, worst first, and you will make every change yourself.

You're practising: Getting a risk list from your AI without it deleting, revoking or changing anything on its own.

The mistake to catch: Agents that can act will act. Make sure nothing is sent, deleted or changed without you.

← Your learning path

1. Open the prompt

Open Grok Bot shared-computer security auditand copy its prompt into your AI. Any AI works.

2. Give it this sample data

Files found:
- .env with a test API key
- private-key.pem in Downloads
- customers_export_oct.csv (names, emails, phone numbers)

Account:
- 2 browser sessions, one unused for 41 days
- 4 bots: one sends email automatically, one unused bot still has payment access
- A routine with a webhook that has no password or token
- 14 failed sign-in attempts this week

Your task: Ask your AI for a risk list, worst first, with the fix for each. It must not change, delete or revoke anything.

3. Paste your AI's answer

Checked in your browser with plain rules. Your answer isn't sent or stored anywhere.

Your learning path · Build your AI staff