Circular No. 53/2026/TT-NHNN: AI-automated payments for individual customers must not exceed VND 100 million per day

Circular No. 53/2026/TT-NHNN of the State Bank of Vietnam, prescribing safety, risk management, and conditions for deploying artificial intelligence systems in the banking sector, was issued on September 30, 2026.

Under Item (ii), Point g, Clause 1, Article 5 of Circular No. 53/2026/TT-NHNN:

(ii) The total value of transactions automatically executed by an artificial intelligence system for a customer in one day must not exceed the limit prescribed in the institution’s internal regulations and, in any event, must not exceed VND 100 million for individual customers or VND 500 million for institutional customers;

Accordingly, the total value of transactions automatically executed by AI for a customer in one day must remain within the institution’s internal limit and must not exceed VND 100 million for individual customers or VND 500 million for institutional customers.

The customer and the institution must enter into an agreement authorizing AI to automatically execute payment transactions. The agreement must clearly specify the scope of permitted transactions, per-transaction and daily limits, the validity period, and conditions for suspension. The customer has the right to terminate the agreement at any time, and the institution must immediately act on the termination request.

When entering into the agreement, the institution must check, cross-check, and verify customer identification information to ensure that the agreement is entered into by the account holder or a representative as prescribed.

If the agreement is entered into electronically, verification must, at a minimum, involve biometric matching combined with another form of authentication under Circular No. 50/2024/TT-NHNN, as amended and supplemented by Circular No. 77/2025/TT-NHNN.

The institution must also provide tools enabling customers to suspend or revoke authorization for AI-automated payments on their own initiative. In addition to the specific payment conditions, Clause 1, Article 5 requires the model to be tested and assessed as meeting quality thresholds, with risk thresholds, warning criteria, and monitoring procedures in place.

The institution must also ensure the ability of humans to intervene and determine whether a transaction was automatically executed by AI or carried out by a human. For high-risk systems, tools must be established to continuously monitor system operations.