Vietnam's central bank issued binding safety and risk-management rules for artificial-intelligence systems used in banking. Circular 53/2026/TT-NHNN was issued on 30 September and takes effect on 15 November, according to the government's legal database. It applies to credit institutions, foreign bank branches and other regulated financial-service providers.
The rules cap payments executed automatically by AI at VND100 million per day for an individual and VND500 million for an organisation. Providers must apply any lower internal limit and obtain an agreement specifying the permitted transactions, limits, duration and suspension conditions. Customers must be able to terminate the authority and have that request acted on immediately.
HIGH-RISK SYSTEMS FACE EXTRA CONTROLS
AI payment systems are classified as high risk at lower transaction thresholds than the absolute daily caps. BIDV's information service said those thresholds include more than VND10 million in a single individual transaction or VND20 million daily, and VND50 million or VND100 million respectively for organisations.
The circular requires testing, monitoring thresholds, warning criteria and human intervention for relevant systems. It also states that AI does not replace human authority or responsibility, leaving providers legally accountable for decisions made or supported by their systems.
EXISTING SYSTEMS RECEIVE A TRANSITION PERIOD
Institutions must manage model, data, cybersecurity, outsourcing and change risks rather than treating third-party technology as a transfer of responsibility. The framework therefore links automated payment limits with broader governance throughout an AI system's life cycle.
The first milestone is the circular's 15 November commencement. AI systems already operating before that date must meet the safety, risk-management and deployment requirements by 1 March 2027, giving supervisors and banks a defined transition window.