Security · Data protection

Malaysia's amended PDPA: what changes for teams that build and run software

Breach notification, data protection officers and direct duties for processors came into force in 2025. If you build or operate systems that hold personal data, several of these are engineering requirements.

Pikabyte Engineering 4 min read

72 hours

the reported window for notifying the Commissioner of a significant data breach

The Personal Data Protection (Amendment) Act 2024 brought the most significant changes to Malaysia’s data protection law since it was passed. The provisions came into force in stages during 2025. Law firms have published detailed guides, and you should read one and take advice. This article is about the parts that land on an engineering team’s desk.

What changed, in brief

According to published legal summaries:

From 1 April 2025

  • Data processors have direct obligations. A company that processes personal data on behalf of another, which describes most software vendors and hosting arrangements, is now directly subject to the security principle.
  • Higher penalties. The maximum fine for breaching the data protection principles rose from RM300,000 to RM1,000,000, and the maximum prison term from two years to three.
  • Changes to cross-border transfer rules and updated definitions.

From 1 June 2025

  • Mandatory data breach notification. A breach that causes or is likely to cause significant harm must be notified to the Commissioner. Guidance summarised by practitioners describes a 72-hour window, with further details permitted in phases afterwards, and treats a breach affecting more than 1,000 people as significant by scale.
  • Data protection officers. Controllers and processors that meet the criteria must appoint at least one, and register the appointment.
  • Data portability. Individuals can ask for their data to be transmitted to another controller, subject to technical feasibility.

Check the Act, the regulations and the Commissioner’s guidelines for the precise thresholds. What follows is how we translate them into systems.

1. You cannot notify what you cannot detect

A 72-hour clock starts when you become aware of a breach. The engineering question is how quickly you would become aware, and how quickly you could say what was affected.

That needs:

  • An audit trail of access to personal data, at application level and at infrastructure level, stored where an intruder cannot erase it.
  • Alerts that reach a person: unusual access, unusual spend, failed sign-in spikes.
  • An inventory. Which tables and buckets hold personal data, of what kind, about roughly how many people. Without it you cannot assess scale.

In a cloud compromise we responded to, the first questions were exactly these: what did the attacker’s credentials touch, and when. An organisation-wide audit trail is what answers them.

2. Vendors are now on the hook

If you build or host a system for a client, you are likely a data processor with direct security obligations. Practical consequences:

  • Tenant isolation must be real. In our multi-tenant platforms every record carries its organisation, every query is scoped to it, and automated tests fail the build if a new endpoint forgets.
  • Least privilege for your own staff. Support access should be read-only by default, time-limited, and logged.
  • Know your sub-processors. Every third party that sees the data: cloud provider, email service, and now AI model providers.

3. AI features are data transfers

Sending a customer’s document to a hosted AI model is disclosing personal data to a third party, often in another country. Before building it:

  • Identify which fields actually need to be sent, and send only those.
  • Know the provider’s region and retention terms.
  • Record which data went to which provider.
  • Tell the client. It belongs in the contract and the privacy notice.

4. Design for portability and deletion

A request to export or delete one person’s data should be an operation, not a project. That is easy if considered from the start and expensive later:

  • Keep personal data in identifiable places, linked by a stable identifier.
  • Provide an export in a common format.
  • Know your retention periods and enforce them with scheduled jobs.

5. Remove the usual causes

Most breaches at smaller companies are not sophisticated. They come from a short list:

  • Long-lived credentials in code, chat or laptops. Replace with single sign-on and short-lived access.
  • Databases reachable from the internet. Make them private.
  • Production data copied to test environments or spreadsheets. Use synthetic data.
  • Shared accounts. One identity per person, so the audit trail means something.

We publish the baseline we use for cloud accounts.

A checklist for your next system review

  • Can we list every store of personal data and what it contains?
  • Would we know within a day if someone accessed it who should not?
  • Could we state what was affected within 72 hours?
  • Is every tenant’s data isolated, and is that tested?
  • Which third parties, AI providers included, receive personal data?
  • Can we export and delete one individual’s data on request?
  • Who is our data protection officer, and do engineers know how to reach them?

A team that can answer these is in a good position whatever the precise thresholds turn out to be.

Sources

This article is general information for engineering teams and is not legal advice.

Written by

Pikabyte Engineering

Pikabyte builds custom business systems, applied AI and secure cloud infrastructure from Malaysia. Talk to us.

Start a conversation

Want this applied to your business?

Tell us what is slowing the business down. You will talk to an engineer, not a salesperson, and get a straight answer on whether software is the right fix.