The short answer
Using AI with company data is legal in Malaysia, and for most work it is safe enough — if you use business accounts rather than personal ones, keep customer and staff personal data out of prompts unless you have a reason to put it there, and keep the most sensitive material on AI that runs inside your own office. What is genuinely risky is what most companies do today: staff pasting customer lists, salary tables and signed contracts into personal ChatGPT accounts, with nobody knowing it happens.
What changed in Malaysian law
The Personal Data Protection Act was amended in 2024, with the main provisions in force through 2025. Three changes matter for AI.
- Penalties rose sharply — up to RM1 million in fines and up to three years' imprisonment.
- Breach notification is mandatory — the Commissioner must be told within 72 hours, and affected individuals shortly after. Every outside service holding your data is another party you must chase inside that window.
- Data processors are now directly liable, and cross-border transfers are judged on risk rather than a fixed country list. Sending personal data to an overseas AI service is a transfer you have to be able to justify.
A separate AI Governance Bill went through public consultation in 2026 and is expected to be tabled by early 2027, with risk tiers and a central regulator. Nothing to comply with yet, but it is coming.
Does ChatGPT train on what you type?
It depends entirely on which account you use.
- Personal free and Plus accounts: conversations may be used to improve models unless the user turns that off themselves. Staff rarely do.
- Business, Team and Enterprise plans from the major providers: company data is not used to train models by default. That is the line in their business terms, and it is the main reason to move your team onto company accounts.
- Private AI on your own machine: nothing leaves the building, so the question does not arise.
"Not used for training" is not the same as "never stored". Business plans still keep conversations for a period so users can see their history, and staff with access can read what colleagues typed. Treat an AI account like a shared drive: fine for working material, wrong place for a customer database.
The practical test before you paste anything
Give your staff one rule that is easy to remember: would you be comfortable if this text appeared in an email to a supplier by mistake? If not, it does not go into a cloud AI tool.
In practice that means:
- Usually fine: product descriptions, marketing copy, meeting notes without names, public specifications, general questions, code.
- Strip identifying details first: customer emails and enquiries, supplier quotations, CVs, complaints — the question is usually about the content, not the person, so remove names, IC numbers and phone numbers.
- Keep off cloud AI entirely: patient records, full customer or staff databases, signed contracts under confidentiality terms, financial statements before release, drawings and formulations, anything covered by an NDA.
Seven things to put in place this month
- Move everyone onto company accounts. A business plan is roughly US$20–25 per user a month. Cheaper than one data incident, and you can see who has access.
- Write a one-page AI usage policy. What staff may use, the three data categories above, and who to ask when unsure. One page that people read beats twenty that they don't.
- Train the team once, properly. Most leaks come from not knowing, not from bad intent.
- Decide where personal data is allowed to be processed, and write that decision down. This is what a regulator will ask to see.
- Sign a data-processing agreement with any vendor who builds or runs AI workflows for you. Since 2025, processors carry direct obligations.
- Have a breach plan that fits 72 hours. Who notices, who decides, who notifies. Rehearse it once.
- Appoint a data protection officer if you meet the thresholds — broadly, large volumes of personal data, sensitive or financial data, or regular systematic monitoring. Check your position with your legal adviser.
Three levels of protection, and which one you need
- Level 1 — business cloud AI. Company accounts, no training on your data by default, plus a policy and redaction rules. Right for most trading, distribution and manufacturing work.
- Level 2 — hybrid. Personal details are removed or processed locally, and only clean content goes to cloud models. Right for HR-heavy firms, finance teams and companies holding a lot of customer data.
- Level 3 — private AI in your office. Nothing leaves the building, and it can run offline. Right for law firms, accounting firms, clinics and manufacturers with sensitive designs. See private AI for what that costs.
The mistake we see most often
Companies spend weeks debating whether AI is safe while their staff quietly use personal accounts every day. The unmanaged use is the actual exposure. Moving the team onto company accounts with a one-page policy takes an afternoon and removes most of the risk — before any project starts.
Sources
- PDPA amendments, penalties, processor liability and cross-border rules: Mayer Brown
- Breach notification and data protection officer guidelines: DLA Piper
- AI Governance Bill consultation: Baker McKenzie
- Business-plan data handling: OpenAI business data, Claude plans
This article is general information from an AI implementation provider, not legal advice. Have your legal adviser review your policies, consent notices and data protection officer position.