Guidance that small firms have not read
Ask the owner of a 40-person firm about AI governance and the usual answer is that it is a matter for banks and multinationals. Meanwhile their staff use AI tools every day with company and customer information.
Both Malaysia and Singapore have published guidance that applies to them. Malaysia's Ministry of Science, Technology and Innovation published the National Guidelines on AI Governance and Ethics on 20 September 2024. The guidelines are voluntary and set out seven principles: fairness, reliability with safety and control, privacy and security, inclusiveness, transparency, accountability, and the pursuit of human benefit and happiness.
Singapore's Model AI Governance Framework for Generative AI, published in 2024, sets out nine dimensions to be considered together. Both documents are thoughtful and long. A small company needs a way to act on them without hiring a compliance team.
Why voluntary guidance still matters
It is tempting to file voluntary guidance under later. I think that is a mistake, for three practical reasons.
A larger customer can reasonably ask a supplier how it uses AI, and a clear answer builds trust. Existing law already applies to much of what AI tools touch, including personal data and consumer protection. And habits formed now are cheap to set, while habits corrected after an incident are expensive.
There is also a plain management argument. A firm that does not know where AI is used in its own work cannot judge whether it is helping.
Six things a small firm can write down
The principles become manageable when they are turned into a short record. Six items cover most of the ground.
Keep a list of where AI is used in the business, including the tools staff adopted themselves. Name one person responsible for each use. State which information must never be entered into an AI tool, with customer personal data at the top.
Mark the outputs that a human must check before they are sent or acted on. Decide when you will tell customers that AI was involved. Agree who is told when something goes wrong, and what happens next.
Each item connects to the published principles. The list of uses and the named owners serve accountability. The data rule serves privacy and security. Human checks serve reliability, safety and control. Telling customers serves transparency.
What the page looks like in practice
Take a fictional 40-person distributor. Its record lists four uses. Sales staff use a chat assistant to draft customer emails. Finance uses a tool to extract figures from supplier invoices. Marketing uses an image generator for social posts. One manager uses a transcription tool for meeting notes.
Against each use the page names an owner. The data rule says no customer identity numbers, bank details or staff records go into any tool. Emails to customers and extracted invoice figures are checked by a person before use. Marketing images are labelled as AI-generated where a customer might otherwise be misled. Incidents go to the operations director the same day.
That is the whole document. It answers most of the questions a customer, an auditor or a new employee would ask.
Firms that operate in both countries do not need two records. The Malaysian principles and the Singapore framework point in the same direction on accountability and the handling of data, so one honest page serves both.
Where firms go wrong
The common failure is a policy copied from a large company. It runs to many pages, nobody reads it and practice carries on unchanged. A one-page record that is true is worth more than a manual that is ignored.
The second failure is a ban. Prohibiting AI tools feels safe and drives use onto personal phones, where the firm can see nothing. Approving one or two tools with clear rules gives better control.
The third is treating governance as a one-off. Tools and uses change every few months. Review the record each quarter, in a meeting that takes half an hour.
Proportionate and written
This is operating guidance from a practitioner and no substitute for legal advice. Firms in regulated sectors should take their own counsel.
For everyone else, the standard to aim for is proportionate and written. Know where AI is used, know who owns it and know what must not go in. That fits on a page and can be done in a week.
The first draft will be incomplete, and that is acceptable. Staff will mention tools the owner had not heard of. Add them to the list without blame, because the purpose of the record is to see what is happening.
Governance of this kind is part of how I approach every advisory engagement, because an AI workflow that nobody owns will fail on its own schedule. Start the page now, while the list of uses is still short enough to write from memory.