The State Bank of Vietnam has classified AI-automated transactions exceeding VND10 million (US$380) per transaction or VND20 million ($760) per day for individual customers as high-risk under new banking regulations.
The requirements are stipulated in Circular 53/2026, which governs the safety, risk management and deployment of artificial intelligence (AI) systems in the banking sector and has recently taken effect.
Under the circular, transactions are classified as high-risk if an AI system automatically executes payments exceeding VND10 million ($380) per transaction or VND20 million ($760) in total per day for an individual customer.
For institutional customers, the thresholds are VND50 million ($1,900) per transaction and VND100 million ($3,800) per day.
AI systems that automatically make lending decisions involving VND50 million ($1,900) or more are also classified as high-risk.
These thresholds determine the risk classification of AI systems and the level of supervision banks must apply.
The circular also caps AI-automated transactions at VND100 million ($3,800) per day for each individual customer and VND500 million ($19,000) for each institutional customer, subject to banks’ internal limits.
The State Bank of Vietnam said banks increasingly use AI for credit scoring, loan assessments, payment fraud detection, automated advisory services and electronic know-your-customer (eKYC) procedures.
However, AI also poses risks involving cybersecurity, data protection and model bias.
Banks and customers must sign an agreement before AI systems can automatically execute payments.
The agreement must specify authorized transaction types, per-transaction and daily limits, its validity period and conditions for suspending the authorization.
Banks must verify that signatories are account holders or their legally authorized representatives.
For electronic agreements, banks must use biometric authentication combined with at least one other authentication method as required by regulations.
Customers may terminate the agreement at any time, and banks must immediately process their requests.
Banks must also provide tools for customers to suspend or revoke AI payment authorization.
Banks must arrange human reviews when customers complain about automated decisions.
The review outcome will serve as the final basis for resolving each complaint.
Banks remain liable for AI decisions
The circular requires banks to establish mechanisms for human supervision and intervention in AI operations.
It states that AI does not replace human authority or responsibility, and institutions remain legally accountable to customers for decisions made by AI systems, or decisions supported by AI, that they rely on to provide services.
Institutions cannot cite the technical characteristics or autonomy of AI systems, or results provided by third parties, to refuse, waive or reduce their legal, management or compensation responsibilities, except where exemptions are permitted by law.
AI systems that were officially operational before November 15, 2026 must meet all new requirements by March 1, 2027.
The regulations establish a common framework for AI applications in banking, with requirements aimed at ensuring system safety and protecting customers’ interests.
The Ky – Anh Hong / Tuoi Tre News



