New clauses in Wells Fargo's updated Online Access Agreement stipulate that when a user authorizes a third-party AI agent to access their account using online banking credentials, any instructions and transactions initiated by the agent are deemed to be the user's own actions, with the user bearing the risk. The bank does not commit to monitoring the agent's behavior and may restrict access by specific tools at any time. The core of this clause lies in the fact that current technology cannot distinguish between operations performed personally by the user, those performed by an authorized AI agent, and those performed by malicious programs. The bank leverages the interpretive latitude of the rules accompanying the U.S. Electronic Fund Transfer Act to solidify, through contract, the legal relationship of "the agent is the user," filling the liability-determination gap during a period of regulatory blank space. At present, the U.S. banking industry has formed two divergent paths—blocking and opening up. Wells Fargo has chosen open access with user liability placed upfront, aiming to capture the traffic scenarios of AI agent payments. Mature blueprints already exist for scoped delegated authorization technology solutions, but they have not yet been implemented due to factors such as the high cost of retrofitting banks' legacy core systems, the absence of mandatory regulatory requirements, and industry standards still being in draft form. By comparison, domestic governance starts from the premise that licensed institutions bear primary responsibility, and through dedicated self-regulatory conventions clarifies requirements such as the KYA mechanism, authorization boundaries, and risk control, establishing the underlying logic that "risk follows the service, and liability follows the revenue"—forming a marked contrast with Wells Fargo's transitional arrangement of shifting risk to users.