Permissions and Safety for Assistants
Decide what an assistant may read, draft, send and change, set approval steps, protect secrets and review what it did.
This course is being finished. Tell me when it is ready
Who it is for
anyone who gives an AI assistant access to email, files, calendars, accounts or customers.
What you will be able to do
- Sort every task into three levels: read, draft and act, and decide the level for each
- Keep passwords, keys and private data out of places where they do not belong
- Set approvals, spending limits and boundaries the assistant cannot cross
- Review what the assistant did, fix problems and know how to switch it off
- Check your own work with this test: you can stop the assistant in a minute and you know what it did last week.
The programme
Module 1: Read, draft, act and Secrets and private data
Lessons 1 to 2: Read, draft, act; Secrets and private data.
- Read, draft, act. An assistant can do three kinds of things. It can read: see your files, mail or calendar.
Practice: Sort ten of your tasks into the three levels. - Secrets and private data. Never paste passwords, API keys, card numbers or identity numbers into a chat. A secret that appears in a conversation, a log or a file is exposed.
Practice: Review what your assistant can access. Remove anything it does not need.
Module 2: Approvals and limits and Logs, review and shutdown
Lessons 3 to 4: Approvals and limits; Logs, review and shutdown.
- Approvals and limits. Write the rules down, plainly. Examples: no messages to customers without my approval; no spending over a set amount; no deleting files, only moving to an archive folder; no sharing documents outside the company.
Practice: Write your rules page and set one permission in the tool. - Logs, review and shutdown. Keep a log of what the assistant read, drafted and did. Review it weekly at first: anything surprising?
Practice: Set up a log, test the off switch and schedule the review.
What you receive
- 4 lessons from "Read, draft, act" to "Logs, review and shutdown", each with two worked examples and a practice task
- Copy-ready scripts for your AI helper, one per lesson topic
- Workbook sheets: task levels, access list, rules page, technical limits, weekly log review, incident note, quarterly review
- Kailini AI as teacher and guide for "Permissions and Safety for Assistants", available at any hour
How Kailini AI teaches you
As teacher. Kailini AI teaches "Read, draft, act" through to "Logs, review and shutdown" in plain language, adapts the worked examples to your own situation and explains why a result went wrong.
As guide. For the practice tasks, such as "Sort ten of your tasks into the three levels", it reads your drafts, points out anything unclear or risky and checks your self-check with you.
The result
By the end of the course you can decide what an assistant may read, draft, send and change, set approval steps, protect secrets and review what it did.
Custom Agent continuation
After the course, a Kailini AI Custom Agent can be set up to follow your permission list: read-only first, drafts for approval, a log you can review and a stop switch. It says it is an AI and never acts beyond the permissions you set. This is optional and separate. See the Kailini AI Custom Agent.
Honest note
Your AI assistant can make mistakes, so you stay in charge of what it may see, say and do. Results vary from person to person, and nothing in this course is a promise that any setup removes all risk. This course teaches a safety method. It does not promise that any setup is risk free. Responsibility for what an assistant may do stays with you.
These courses give methods and examples. We do not promise results or earnings; what you achieve depends on your offer, your effort and your market.
Questions and answers
Who is this course for?
anyone who gives an AI assistant access to email, files, calendars, accounts or customers.
How is it taught and how long does it take?
4 short lessons of about 15 to 20 minutes each, from "Read, draft, act" to "Logs, review and shutdown". Each of the 4 lessons in "Permissions and Safety for Assistants" closes with its own practice task and self-check question. Workbook sheets for this course: task levels, access list, rules page, technical limits. Plan for about 1.5 hours in total.
Will this make me money?
No promise is made. Nothing in this course is a promise of results.
Do I need a Kailini AI Custom Agent?
No. "Permissions and Safety for Assistants" stands on its own, and every task is yours to do. Later, an optional Custom Agent can be set up to follow your permission list: read-only first, drafts for approval, a log you can review and a stop switch.
Can I try something first?
The programme above names the practice task of each of the 4 lessons. For "Permissions and Safety for Assistants" the first one is: Sort ten of your tasks into the three levels.
Is any setup completely safe?
No. The goal is to limit harm and notice problems early.