Central bank enhances oversight of AI in banking operations

Central bank enhances oversight of AI in banking operations

The State Bank of Vietnam is drafting a circular to regulate safety, risk management and implementation conditions for AI systems in banking, aiming to clearly define management requirements for AI adoption in the sector.

Participants at a seminar on AI and data governance in banking in Hà Nội. — Photo courtesy of the organiser

The State Bank of Vietnam (SBV) will classify AI systems in the banking sector based on risk levels to strengthen supervision of the technology's application in banking operations, officials said at an event in Hà Nội on Wednesday.

Speaking at a seminar on AI and data governance in banking operations, Vietnam Banks Association General Secretary Đào Minh Tú said the global finance and banking sector is undergoing a pivotal shift, moving from the era of digitalisation to the era of AI, and ultimately toward an AI-native banking model.

AI is not limited to chatbot applications or the automation of repetitive tasks, but plays a direct role in core areas of banking operations. It contributes to solving business challenges ranging from trade finance and cross-border payments to asset management.

However, Tú noted, according to international experts and practical experience, data quality determines AI’s reasoning capabilities. When data is fragmented, inconsistent or lacks verifiable provenance, algorithmic models inherit and amplify existing errors. This causes biased outputs that infringe upon customer rights and expose banks to severe non-financial risks.

The cost of an inefficient data governance framework is also exorbitant, consuming an average of 30 per cent of an organisation's total working time in unproductive activity.

"We must shift our mindset from passive, compliance-based policy-making to an AI-enhanced data governance model. The current race in digital banking is no longer about algorithms or the scale of technology investment, but it is a race defined by data quality, comprehensive governance capabilities and digital trust from customers," Tú said.

Hoàng Minh Tiến, deputy director of the SBV’s Information Technology Department, told attendees at the event that AI is being widely applied across the banking sector, with uses ranging from credit scoring, loan appraisal, fraud detection and anti-money laundering to customer service chatbots, eKYC and electronic authentication.

However, alongside these benefits, AI also presents various risks concerning information security, model integrity, data handling and accountability in decision-making processes.

According to Tiến, the SBV is drafting a circular to regulate safety, risk management and implementation conditions for AI systems in banking, aiming to clearly define management requirements for AI adoption in the sector. A key aspect of the draft is the classification of AI systems based on risk levels into three categories: high, medium and low risk.

For high-risk AI systems, the draft aligns with Decision 33/2026, which requires tight controls. In the banking sector, two types of AI systems fall into this category: those that automatically execute electronic transactions and those that automatically make credit granting decisions.

AI systems are classified as high risk when they automatically execute payment transactions ranging from VNĐ10 million to under VNĐ100 million for individual customers, or from VNĐ50 million to under VNĐ500 million for institutional customers. These systems are permitted to make automated decisions only within these value thresholds; automated decision-making is not allowed for transactions exceeding these limits.

Under the draft, the classification of medium- and low-risk AI systems is to be conducted in accordance with the Law on Artificial Intelligence and Decree 142/2026. The Ministry of Science and Technology will also develop support tools for entities to reference during the AI ​​system risk classification process.

Aside from risk classification, the draft places special emphasis on a requirement for human oversight.

AI systems may only make automated decisions when they fully meet conditions regarding model quality, risk thresholds, alert criteria, monitoring procedures and the capability for human intervention.

Banks’ monitoring boards must have the authority to adjust or halt AI decisions when necessary. Furthermore, the boards must maintain an emergency shutdown mechanism for instances involving emerging risks or serious incidents. 

Bizhub