AI in Medical Training

Patient Data and AI in Malaysia: What Is Allowed

What Malaysian clinicians may put into AI tools: MMC clause 3(6), health data as sensitive data under the PDPA, cross-border transfers, de-identification.

  • By
  • Updated
  • 4 min read
  • Facts checked
Short answer

Patient data must not be shared on any AI system without the patient's consent (MMC guideline clause 3(6)(b)), and the doctor must ensure the tool has robust encryption and access control. Health information is also sensitive personal data under the PDPA, which generally needs explicit consent. In practice: no identifiable data in public tools, and approved institutional systems only.

Part of the guide: AI in Healthcare in Malaysia: Courses and Rules for Doctors, Nurses and Hospitals (2026)

On this page
  1. What does the MMC guideline say?
  2. What does the PDPA add?
  3. Do the rules differ in government hospitals?
  4. What counts as de-identified?
  5. What is a practical rule set for a department?
  6. How do the rules apply to everyday scenarios?
  7. How do you explain it to patients?
  8. What should you do next?
  9. Sources
  10. Frequently asked questions

The short rule for clinicians: identifiable patient data does not go into AI tools unless the patient has consented and the tool sits inside a system your institution has approved. The detail behind that rule comes from two sources, the MMC guideline and the Personal Data Protection Act.

Checked on 22 October 2026. This is a summary for clinicians. Legal questions need a lawyer or your hospital's data protection lead.

What does the MMC guideline say?

Clause 3(6) has three parts:

  • (a) practitioners must safeguard patient information;
  • (b) "Patient data must not be shared on any AI system without the patient's consent";
  • (c) practitioners must ensure the tool has a robust encryption system and access control protocols to prevent data breaches and re-identification.

Clause 3(5)(b) adds that patients should be informed beforehand how AI is used in their care, the safeguards against errors and data breaches, and any conflicts of interest.

What does the PDPA add?

Under the Personal Data Protection Act 2010, information about a person's physical or mental health or condition is sensitive personal data, and processing it requires explicit consent unless a listed exception applies. Section 40(1) includes an exception for medical purposes where the processing is done by a healthcare professional or a person owing an equivalent duty of confidentiality. That exception concerns the clinician's own processing. Passing data to a commercial AI vendor raises separate questions, so ask your data protection lead. The 2024 amendments (Act A1727) were phased in during 2025: biometric data was added to the sensitive category from 1 April 2025, and mandatory breach notification and data protection officer rules followed from 1 June 2025 (amendment Act A1727). Cross-border transfers are governed by section 129, explained by the Commissioner's guideline of 29 April 2025.

Question MMC guideline PDPA
Who is bound? Registered medical practitioners Those processing personal data in commercial transactions
Consent Patient consent before sharing data on an AI system Explicit consent for sensitive data, with exceptions
Security Encryption and access control (3(6)(c)) Security principle
Breach Report AI complications to developers and regulators (3(14)) Notification duties from 1 June 2025
Overseas processing Not addressed directly Section 129 conditions

Do the rules differ in government hospitals?

Section 3(1) of the PDPA (Act 709) says the Act "shall not apply to the Federal Government and State Governments", so the Act itself does not govern a ministry hospital's own processing. A peer-reviewed review draws the same conclusion for public healthcare. Government hospitals follow Ministry of Health policy, and the MMC guideline still binds the doctors working there. Do not read the PDPA's limited reach as permission to paste patient data into a public tool.

What counts as de-identified?

Removing a name is not enough. Names, IC numbers, addresses, telephone numbers, exact dates, hospital registration numbers and photographs are obvious identifiers. In small communities or for rare conditions, a combination of age, location and diagnosis can be enough. If your hospital has a de-identification standard, use it. Otherwise use fully constructed cases for teaching and practice.

What is a practical rule set for a department?

  1. Approved list. Name the tools approved for patient data and the ones that are not.
  2. No identifiers in public tools. State it in writing.
  3. Consent route. Define when a patient is told that AI is used.
  4. Contract check. For any vendor, confirm data location, retention, training use and access controls.
  5. Incident route. Say who is told when something goes wrong.
  6. Training record. Everyone who uses AI has been trained (clause 3(4)).

How do the rules apply to everyday scenarios?

Scenario Allowed? Why
Ask a public chatbot to explain a drug class Yes No patient data
Paste a patient's name and diagnosis into a chatbot No Identifiable patient data without consent, clause 3(6)(b)
Paste a de-identified summary of a rare condition Doubtful Re-identification risk; clause 3(6)(c)
Use a hospital-approved dictation tool Yes, if approved and consent and policy are met Institutional control of data
Upload an ECG photo to a free app No Image is patient data
Build a teaching case with invented details Yes No real patient
Use real, consented patient story in a lecture through a public AI tool Only with specific consent and an approved route Consent must cover this use
Ask a colleague overseas to run your data through their AI Treat as a transfer; get advice PDPA section 129

How do you explain it to patients?

Clause 3(5)(b) lists what patients should be told before the consultation, where appropriate: how AI is being used in their care, the basis of the result, known risks, safeguards against errors and data breaches, the right to know if the tool fails, and any conflicts of interest. A plain version is: "We use a tool to help with [task]. I check everything it produces. Your information stays inside [the hospital's system]. You can ask me not to use it." Adapt to your institution's wording, and make sure the sentence is true before you say it.

What should you do next?

Our AI course for hospital teams turns these points into a department policy draft. For tools, see Claude vs ChatGPT for doctors. For the wider legal frame, read PDPA and AI integration in Malaysia and the AI in healthcare guide for Malaysia. To plan, book a free 30-minute call.

Sources

(checked November 2026)

General information, not medical or legal advice.

Frequently asked questions

Is removing the patient's name enough to use AI?

Often not. Rare conditions, dates, locations and other combined details can identify someone. De-identification needs a proper method and your institution's approval, and when in doubt the data stays out.

Does the PDPA apply to government hospitals?

Section 3(1) of the PDPA says the Act shall not apply to the Federal Government and State Governments, and the Act covers personal data in commercial transactions. Government hospitals work under Ministry of Health rules, and the MMC guideline still binds their doctors. Confirm with your hospital's legal team.

Can patient data be processed by an AI company overseas?

Section 129 of the PDPA restricts transfers of personal data outside Malaysia unless a listed condition is met, and the Commissioner issued a cross-border transfer guideline on 29 April 2025. Get legal advice before any transfer.

What must happen if there is a data breach involving an AI tool?

The 2024 PDPA amendments introduced mandatory breach notification, phased in from 1 June 2025. MMC clause 3(14) also requires doctors to report untoward complications from AI use to the tool's developers and relevant regulators.

Can I use patient images or ECGs in a public AI tool?

Images and traces can identify a patient and are patient data. The MMC consent rule applies, and public tools do not meet the encryption and access control expectation in clause 3(6)(c).