Commercial banks increasingly rely on artificial intelligence not only to score credit but also to negotiate, conclude and self-execute contracts through smart-contract code, from automated loan disbursement to algorithmic trade-finance and treasury agreements. When such a contract causes damage, whether through an erroneous disbursement, an unauthorised transfer or a flawed automated negotiation, existing civil-liability doctrine struggles to answer a basic question: who is legally responsible, the bank that deployed the system, the vendor that built it, or the customer who accepted it? This article examines the theoretical nature of AI-concluded and AI-executed smart contracts, analyses the doctrinal basis for allocating civil liability for the damage they cause, and compares the approaches of the European Union, the United States, the United Kingdom and Singapore with the current legal framework of Vietnam. The article finds that Vietnamese law, despite recent legislative progress through the Law on Electronic Transactions and the Law on Artificial Intelligence, still lacks a clear rule attributing the acts of an autonomous banking system to a specific legal person and lacks a risk-based liability regime suited to opaque, self-executing code. The article proposes concrete lessons for Vietnam, including a statutory attribution rule, a rebuttable presumption of fault for high-risk banking AI, mandatory audit trails, and contractual risk-allocation and insurance mechanisms between banks and technology vendors.