中文sales@inspirentech.co
Home/Insights/Malaysia's AI Governance Bill: what companies in Malaysia should do before it becomes law
Insights · 5 min read

Malaysia's AI Governance Bill: what companies in Malaysia should do before it becomes law

What Malaysia's proposed AI Governance Bill covers, the risk tiers, the rules that already apply, and a six-step checklist to prepare your team.

On 29 October 2026, Malaysia's Digital Minister Gobind Singh Deo said the country's AI Governance Bill is complete, will go to Cabinet soon and is planned for tabling in the Dewan Rakyat by early 2027. For companies that use AI in Malaysia, that gives roughly a few months, not years, to get organised. The good news: most of the preparation is good practice you would want anyway.

What is Malaysia's AI Governance Bill?

It is a proposed cross-sectoral law that would set out how organisations that develop, provide or use AI systems must manage the risks those systems create. It is designed to sit alongside existing laws such as the Personal Data Protection Act (PDPA), not replace them.

The details below come from the public consultation the National AI Office (NAIO) ran in July 2026 and from the minister's statements since. The final text has not been published yet, so some points may change before the bill becomes law.

What we know so far

  • Risk-based. Obligations depend on the potential harm a system can cause, not just the industry it is used in. The minister told Parliament on 5 October that high-risk systems will face stricter safety controls, testing and human oversight.
  • Developers and deployers. The consultation paper separates those who build or materially change an AI system (developers) from those who decide where and how it is used (deployers). A company can be both. Most businesses using Claude, ChatGPT, Microsoft Copilot or Google Gemini at work would mainly be deployers.
  • Duties scale with control. Obligations would follow each party's role, how much control it has over the system, and the risk the system poses across its life.
  • Incident reporting. The bill is expected to include procedures for reporting and handling AI incidents. The consultation paper proposed covering near misses as well as actual harm.
  • A central AI authority. The paper proposed a body covering AI safety, investigation and enforcement, and enablement, including guidance, training and a regulatory sandbox, working with existing sector regulators.
  • Reach beyond Malaysia. The consultation paper proposed covering systems used in Malaysia or used by deployers established in Malaysia, wherever the system is hosted. Regional companies with Malaysian operations should watch this point.

The proposed risk tiers

The consultation paper set out three tiers. Treat this as the direction of travel rather than final law.

TierWhat it coversWhat it would mean
UnacceptableAI developed or deployed with intent to cause harmProhibited
High riskNo intent to harm, but a real risk of harm (for example, to safety or legal rights)Risk assessment, documentation, internal controls, human oversight, monitoring and mitigation
Low riskNo foreseeable material harm, such as drafting and summarising everyday documentsLighter duties; responsible use still expected

Risk would be judged on likelihood, severity and scale, and how long-lasting or reversible the harm is. A chatbot that drafts marketing copy and a tool that screens job applicants or sets credit limits are likely to land in very different places.

What already applies today

Companies do not need to wait for the bill. Three things already shape how AI should be used in Malaysia:

  1. PDPA amendments. Since 1 June 2025, many organisations must appoint a Data Protection Officer, and personal data breaches must be reported to the Commissioner within 72 hours. Anything staff paste into an AI tool is still personal data processing.
  2. The PDP Commissioner's April 2026 guidelines. Guidelines on data protection impact assessments (DPIA), automated decision-making and profiling, and data protection by design. A DPIA is expected for processing more than 20,000 people, or more than 10,000 where sensitive data is involved, and for high-risk uses including AI. Where automated decisions significantly affect people, privacy notices should say so.
  3. The National Guidelines on AI Governance and Ethics (AIGE). Voluntary guidelines launched in September 2024 with seven principles: fairness; reliability, safety and control; privacy and security; inclusiveness; transparency; accountability; and the pursuit of human benefit and happiness.

A six-step readiness checklist

  1. List every AI system you use. Include obvious tools (Claude, ChatGPT, Copilot, Gemini) and AI built into HR, CRM, finance and customer service software. Shadow use counts too.
  2. Classify each use by risk. Ask: does this output affect someone's job, credit, health, safety, pricing or access to a service? If yes, treat it as potentially high risk.
  3. Name an owner. Give each system a business owner for outcomes and a technical owner for access, settings and logs.
  4. Make human review real. For high-risk uses, the reviewer needs time, authority and training to disagree with the AI. A rubber stamp is not oversight. Our article on AI and human judgement at work explains why this is harder than it sounds.
  5. Set up incident handling. Decide who staff tell when AI gets something seriously wrong, how it is logged, and how it links to your 72-hour data breach process. Include problems at your AI vendors.
  6. Write it down and train people. A short acceptable-use policy plus a practical briefing does more than a long policy nobody reads.

Who needs to do what

RolePriority now
Leadership and boardAgree who owns AI risk and how it is reported upwards
Legal, compliance and DPOTrack the bill, map it against PDPA duties, run DPIAs for high-risk uses
IT and securityKeep the AI inventory, approve tools, manage access and logs
HR and L&DCheck AI used in hiring and performance decisions; run staff training
Every employeeFollow the policy, protect personal data, check output, report problems

Put it in your AI policy

Your acceptable-use policy is the easiest place to start. It should name approved tools, set data rules, say which tasks need human review, and explain how to report an AI incident. Use our AI acceptable-use policy checklist or generate a first draft with the free AI policy generator, then have it checked against Malaysian law. If your team also operates in Singapore, our guide to Singapore's Model AI Governance Framework covers similar ground.

Where to start

Take the free AI readiness quiz to see where your team stands. For teams in Malaysia and across Asia, our responsible AI and governance training and AI literacy training run live online, so the whole team learns the same rules at the same time.

Frequently asked questions

Is Malaysia's AI Governance Bill law yet?

No. As of 29 October 2026 the bill is complete and going to Cabinet, with tabling in Parliament planned by early 2027. It still needs to pass Parliament and come into force.

Does the bill apply to companies that only use AI tools like ChatGPT or Claude?

Based on the consultation paper, yes in principle: organisations that use AI systems would be deployers, with duties that scale with risk and control. Everyday low-risk use is expected to carry lighter duties.

What counts as high-risk AI in Malaysia?

The final list is not published. The consultation paper described high risk as systems that, without intending harm, still create a risk of harm, judged on likelihood, severity, scale and reversibility. Uses affecting jobs, credit, health or safety deserve the closest look.

What should a Malaysian company do before the bill passes?

List the AI systems you use, classify them by risk, assign owners, make human review meaningful, set up incident handling linked to your PDPA breach process, and train staff on a clear AI policy.

This article is general information for business readers, not legal, financial or technical advice. AI tools and plans change quickly; check current vendor documentation before making decisions.

Keep exploring

Ready to make AI work for your team?

Tell us about your team and goals. We'll reply with a tailored training plan and quote.

Get a Quote