AI governance for Malaysian financial services: trust belongs on the board agenda.

AI adoption in Malaysian finance is already mainstream

In July 2026, Bank Negara Malaysia’s Governor said more than 70% of Malaysian financial service providers had implemented at least one AI application. Use cases are expanding across fraud detection, credit and insurance risk, compliance, operations and customer service.

BNM’s message was equally clear: trust, accountability and professional judgement determine whether that adoption endures. AI is therefore a board, risk and operating-model issue—not only a technology programme.

Use the emerging framework as a control baseline

Malaysia’s banking industry released an AI Governance Framework through the Chief Risk Officers Forum of the Asian Institute of Chartered Bankers, with BNM support. BNM has also revised its Risk Management in Technology policy to strengthen technology and cyber resilience and facilitate secure adoption of advanced technology.

Institutions should align AI-specific controls with existing lines of defence, technology risk, outsourcing, customer information and incident management rather than create a disconnected AI checklist.

Govern the decision path, not only the model

A model score or generated answer becomes consequential through the workflow around it. Record the customer or staff identity, applicable entity boundary, retrieved information, model version, evaluation, override and final human decision.

For agents, separate permission to analyse from permission to transact. A tool call that changes a customer record, communicates externally or triggers a payment should require explicit authority and, where appropriate, named approval.

• Board-approved purpose, risk appetite and accountable owner.
• Independent validation for material models and workflows.
• Entity-aware knowledge and customer-data boundaries.
• Continuous monitoring for quality, bias, security and cost.
• Exportable evidence for audit and supervisory review.

Run a financial AI trust exercise

Select one live AI-assisted decision and reconstruct it from outcome to source. Confirm that reviewers can see the applicable policy, data, model, explanation, override and approval without joining logs manually.

Then repeat the request from another entity or role that should not have access. A controlled denial with complete evidence is a stronger governance result than another successful demo.

◇

In financial AI, the durable advantage is not automation alone—it is automation that preserves trust.