Security · Governance

Your staff already use AI. A one-page policy for a 50-person company

Generative AI is already in everyday use in Malaysian workplaces, and few employees have been trained for it. Banning it does not work. A short, practical policy does.

Pikabyte Engineering 3 min read

12%

of Malaysian employees say they have had enough AI training, in a 2025 EY survey

Whether or not your company has an AI strategy, your company uses AI. EY’s 2025 survey of Malaysian employees, as reported, found generative AI in wide use at work, with users saying it saves them many hours a week. It also found that only 12% felt they had received enough training to use it well, and that 68% said their workload had gone up, not down.

Much of that use happens on personal accounts, with no oversight. The name for it is shadow AI. A sales executive pastes a customer contract into a chatbot to summarise it. An accounts clerk uploads a spreadsheet of salaries to fix a formula. Nobody means harm. Nobody has been told what is acceptable either.

Banning AI tools does not solve this. People who find a tool useful keep using it, on their phones, where you cannot see. What works is a short policy that people can remember.

The risks, without the drama

Confidential data leaves the company. Text pasted into a consumer AI tool is sent to that company’s servers and, depending on the product and settings, may be kept or used to improve its models.

Personal data is disclosed. Customer and employee data is covered by the Personal Data Protection Act. Sending it to an unvetted third party is a compliance problem, and breach notification is now mandatory.

Wrong answers are trusted. AI output is fluent and sometimes false. Malaysian courts have already dealt with lawyers who filed citations to cases that do not exist.

Nobody knows what was AI-written. When something goes wrong, you cannot trace it.

A policy that fits on one page

Adapt this. Keep it to one page.

1. You may use AI tools for work. We want you to. Use the tools on the approved list. If you need another, ask.

2. Never put these into any AI tool that is not on the approved list:

  • Customer names, identity numbers, contact details or account information
  • Employee records, salaries or medical information
  • Contracts, pricing, tenders and anything marked confidential
  • Passwords, access keys or system configuration

3. You are responsible for what you send out. Check every fact, figure, name and reference. “The AI wrote it” is not an explanation.

4. Tell us when AI produced something that goes to a customer, a regulator or a court. Your manager should know.

5. AI does not make decisions about people. Hiring, discipline, credit and complaints are decided by a person.

6. If you make a mistake, tell us quickly. Pasting something you should not have is fixable if we know the same day.

What management has to do in return

A policy that only restricts will be ignored. Three things make it workable.

Provide an approved tool. Pay for a business plan of one mainstream assistant, where the contract says your data is not used for training. It costs little per user and removes the reason to use personal accounts.

Train for an hour. Show real examples from your own work: a good use, a bad use, and how to check an answer. With roughly one in eight employees feeling adequately trained, this is the cheapest improvement available.

Name an owner. One person who keeps the approved list, answers questions and hears about mistakes.

When a chatbot is the wrong tool

If the same task is done with AI many times a day on sensitive material, a chat window is the wrong shape. That is the point to build it into your own system, where the data sent to the model is limited to what the task needs, the output is checked on a review screen, and there is a record of who approved what.

A law firm we work with reads figures from bank statements this way. The model sees one document, returns a fixed set of fields, and a person confirms them. Nobody pastes anything anywhere.

Start this month

  1. Ask your team, without blame, which AI tools they use and for what.
  2. Choose one approved tool and pay for it.
  3. Issue the one-page policy.
  4. Run a one-hour session.
  5. Review in three months.

The aim is not to control every keystroke. It is to make the safe way the easy way.

Sources

This article is general information and is not legal advice.

Written by

Pikabyte Engineering

Pikabyte builds custom business systems, applied AI and secure cloud infrastructure from Malaysia. Talk to us.

Start a conversation

Want this applied to your business?

Tell us what is slowing the business down. You will talk to an engineer, not a salesperson, and get a straight answer on whether software is the right fix.