AI in Malaysia · Financial services

Bank Negara has set out its thinking on AI. Here is what it means if you build for banks

The central bank's discussion paper treats existing technology rules as the baseline and responsible use as the expectation. Vendors and panel firms should read it as a specification.

Pikabyte Engineering 3 min read

77%

of insurers and takaful operators had at least one AI application in use, per BNM's 2024 survey

On 5 August 2025 Bank Negara Malaysia published a discussion paper on artificial intelligence in the Malaysian financial sector and invited feedback until 17 October. It is not a regulation. It is the central bank setting out how it is thinking, which for anyone selling into banks is the next best thing.

Adoption is already well under way. The paper cites BNM’s own 2024 survey of financial service providers: 71% of banking institutions and development financial institutions, and 77% of insurers and takaful operators, had implemented at least one AI application. Reporting on the paper adds a caveat worth keeping in mind. Having one application in use is not the same as having scaled AI across the business.

What the paper says, in plain terms

The existing rules already apply. Commentary on the paper notes that BNM regards its current, technology-neutral framework for technology risk as adequate for now. AI is not a regulatory gap. It is a technology, and the rules for managing technology risk cover it.

Responsible use is the expectation. Fairness, accountability and transparency are the principles institutions are expected to embed.

Concentration is a systemic concern. The paper flags that if the whole sector converges on the same foundation models or the same datasets, risk could be introduced or amplified across institutions at once.

The definition is still open. BNM asked whether a formal, sector-specific definition of AI would help. Industry bodies responded that a uniquely Malaysian definition could fragment regulation across markets.

Read it as a requirements document

If you supply software or services to a financial institution, a law firm on a bank’s panel included, each of those points becomes a question you will be asked.

“Show us the control.” Because existing technology risk rules apply, an AI feature needs the same things any system needs: access control, audit trail, change management, testing, a documented owner. A model bolted onto a process with none of these will not pass review.

“Why did it produce that?” Accountability and transparency mean every AI output that affects a customer needs a trail: the input, the model and version, the output, and the person who accepted it.

“What if the provider fails or changes?” The concentration concern translates into a practical requirement. Do not hard-wire one model provider into the core of a process. Keep the model behind an interface, keep prompts and settings as configuration, and know how you would switch.

“Where does the data go?” Customer data sent to a model provider is data sent to a third party. Know which data, which provider, which region.

What this looks like in a real system

Our banking litigation platform handles recovery work for a bank’s panel firm. Most of it is deliberately not AI.

Interest on a judgment debt is computed by a deterministic engine, to the cent, with separate handling for conventional and Islamic facilities, and exported in the bank’s own statement format. The same inputs always give the same figures, and the calculation is covered by automated tests. A bank can audit that.

AI is used for one narrow task: reading figures out of statements of account. The extracted values are presented to a person for confirmation before anything is calculated from them. The model can be replaced without touching the calculation.

Around it: role-based access configurable per firm, a read-only oversight mode, a register of every generated document, and tenant isolation that is checked by tests on every build.

That division is the pattern we would recommend to anyone building for the sector. Use AI to read. Use code to calculate. Use people to approve. It satisfies the principles in the paper without waiting for the final rules, because it was never relying on the model to be right.

A short checklist

  • Every AI output has a named human owner before it reaches a customer or a ledger.
  • Calculations are deterministic and tested.
  • Each AI output is recorded with its input and the person who accepted it.
  • The model provider can be changed without redesigning the process.
  • You can state which data leaves your environment, and where it goes.

Sources

Written by

Pikabyte Engineering

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

The work behind this article

Platform · Legal · Banking litigation

From a bank's spreadsheet to court-ready documents, with every step on the record.

A multi-tenant platform for law firms that run banking litigation at volume. Monthly debtor lists become typed case records, letters and court documents are generated from the firm's own templates, and post-judgment interest is calculated to the cent in the bank's statement format.

95%
less time spent filing and checking
12,700+
court and recovery documents generated
3,500
debtor accounts taken through civil process
6 weeks
from kickoff to production
Tour Clausework Kek Ooi & Lee Hong · Malaysia

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.