Many business owners assume AI rules are written for banks and telcos. Malaysia's draft AI and Personal Data Protection Framework says otherwise. If your company uses AI on personal data, whether you built the tool or simply subscribe to it, the draft is aimed at you too.
The Personal Data Protection Department (JPDP) opened public consultation on the draft on 5 October 2026, and comments close on 23 October 2026 (PPC Land). This article explains what the draft covers, what is still undecided, and five checks an SME can run now on the AI tools it already uses.
This is general information, not legal advice. For decisions about your own obligations, speak to a lawyer.
What the draft framework is
The draft is written as a Personal Data Protection Guideline, labelled version 1.0, with the issue date still left blank (Rajah & Tann Asia). It is intended as a main reference for data controllers and data processors on how the Personal Data Protection Act (PDPA) applies when AI handles personal data.
Two points matter most for a small business:
- It covers AI you buy, not only AI you build. According to Rajah & Tann's summary, the draft applies to AI developed in-house, bought from a vendor, or used through SaaS, cloud services or an API, including general-purpose AI models built into everyday applications.
- Existing rights still apply. Customers keep their rights to access and correct their personal data when it is processed by an AI system.
Much is still open. Baker McKenzie notes that the final text, the issue date and any transition period have not yet been set (Baker McKenzie). So treat the draft as a strong signal of direction, not a final rulebook.
Why "we just use an off-the-shelf tool" is not a defence
Under the PDPA, the business that decides why and how personal data is used is the data controller. Subscribing to a chatbot or a document-reading service does not hand that role to the vendor. The vendor processes data on your behalf, but the duty to customers stays with you.
The draft also sits on top of rules that already apply. The PDPA amendments brought in a duty to notify the Commissioner of personal data breaches, and the regulator has issued guidance on data protection impact assessments and on automated decision-making and profiling (Tsaaro; Future of Privacy Forum). An AI tool that leaks one customer's details to another is a data breach, whoever wrote the software.
Five checks to run on your AI tools this week
You do not need a compliance team to start. Take each AI tool your business uses, such as a WhatsApp agent, an invoice or receipt reader, or an automation that moves data between systems, and answer these five questions.
1. What personal data goes in?
List what the tool actually sees: names, phone numbers, IC numbers, addresses, medical details, bank details. Many tools see more than they need. A booking bot rarely needs an IC number. An invoice reader does not need staff salaries. Cut what you do not need, because data you never collect cannot leak.
2. Where is it stored, and for how long?
Ask the vendor where conversations, uploaded documents and extracted data are kept, and for how long. Check whether your data is used to train the vendor's models, and whether you can switch that off. If the vendor cannot answer clearly, write that down as a risk.
3. Who is the vendor, and what have they agreed to?
Check that there is a written agreement covering how the vendor handles your customers' data, including security, sub-processors and what happens when you leave. For tools sold through a reseller, find out who the real processor is.
4. How does a customer ask for access or correction?
If a customer asks what you hold about them, or asks you to fix it, can you find their data inside the AI tool and change it? Test this with one real record. If the answer is "we would have to email the vendor and wait", fix the process before you need it.
5. What happens if the bot shows one customer's data to another?
Assume it will happen one day and plan for it. Who notices? Who decides whether it is a notifiable breach? Who contacts the customer? Write the steps on one page, name an owner, and keep logs so you can see what the AI said and to whom.
An example: the five checks on a clinic's WhatsApp booking agent
This is an illustrative example, not a client result.
Picture a private clinic in the Klang Valley that answers WhatsApp enquiries with an AI agent. Patients ask about opening hours, book appointments and sometimes describe their symptoms.
- Data in. The agent needs a name, a phone number and a preferred slot. It does not need an IC number to book, and it should not invite patients to describe symptoms in detail. A short opening message can say so.
- Storage. The clinic asks the vendor where chat histories are kept and for how long, and sets a retention period it can explain to a patient.
- Vendor. The clinic checks that its agreement with the vendor covers health information, which most patients would expect to be handled with extra care.
- Access and correction. Front-desk staff practise finding one patient's chat history and booking record, so a request does not turn into a week of emails.
- Breach plan. The clinic names one person who reviews any report of a wrong message, and keeps the agent's logs so it can see exactly what was sent.
None of this needs new software. It needs an hour, a list and a named owner.
Should an SME comment on the draft?
You can. Feedback goes through the government's Unified Public Consultation Platform, as noted in the consultation coverage (PPC Land). If your business depends on AI tools from overseas vendors, or if parts of the draft look hard for a small team to meet, the consultation is the place to say so. The deadline is 23 October 2026.
Even if you do not comment, use the draft as a checklist. It shows what the regulator is likely to ask about.
How we approach this at Codech
Codech Solutions builds AI agents, document automation and business insights for SMEs in Malaysia and Hong Kong.
When we build AI agents or Document AI for clients, the five checks above are part of the design, not an afterthought. We decide what data the agent really needs, where it is stored, how a record can be found and corrected, and what gets logged. The same applies to automations that move data between your existing systems. You can see how a project runs on our How we work page.
If you would like a second pair of eyes on the AI tools you already use, start with an AI audit. We will go through each tool with you against these five checks and tell you plainly what to fix first.
Frequently asked questions
Does Malaysia's draft AI and Personal Data Protection Framework apply to small businesses?
The draft is aimed at data controllers and data processors that use AI on personal data, and it covers AI that is bought from a vendor or used through SaaS, cloud or an API, not only AI a company builds itself. Size alone does not take a business outside the PDPA. This is general information, not legal advice.
When does the consultation close?
Comments on the draft close on 23 October 2026. The consultation opened on 5 October 2026.
Is the framework final?
No. It is a draft. The final text, its issue date and any transition period have not yet been set.
If our chatbot vendor stores the data, is the vendor responsible instead of us?
Generally, the business that decides why and how personal data is used remains the data controller under the PDPA. Using a vendor does not remove that duty, so you should know what the vendor does with your customers' data and have it in writing.
What is the quickest first step?
List every AI tool your business uses and, for each one, write down what personal data it sees, where it stores that data, and who at your company owns it. That one page already answers most of the five checks.