On August 24, the China Clearing and Payment Association officially issued the "Self-Regulatory Convention for Agent-Based Payment Applications," which took effect immediately upon release. The framework applies to member institutions offering services such as account management, transaction processing, acquiring, and clearing and settlement, with the goal of defining the operational boundaries of artificial intelligence technology within the payment sector.
The convention clarifies that agent-based payment refers to the use of large model technologies by member institutions to interact with digital services and initiate payment instructions on behalf of users, strictly under their authorization and aligned with their genuine intent. As AI capabilities extend deeper into service chains, the security risks associated with autonomous agent decision-making have become increasingly apparent, prompting targeted regulatory measures within the new framework.
The most notable advancement lies in the extension of management protocols. The convention mandates that member institutions build upon traditional Know Your Customer practices by exploring the establishment of a Know Your Agent mechanism. This means that agents integrated into the payment chain must undergo identity recognition and verification, with management structured according to high, medium, and low risk classifications.
On the subject of authorization and spending limits, the convention emphasizes the necessity of defining clear boundaries. Member institutions are required to sign agreements with users that specify transaction caps, designated deduction accounts, payment priority, and validity periods, while also guaranteeing users the right to revoke authorization at any time. Regarding the model where agents autonomously initiate payment transactions, the convention adopts a cautious stance, requiring that such applications be registered with the payment association before any formal deployment.
In terms of risk control, the convention integrates algorithm and model risks into the comprehensive risk management system to prevent financial losses stemming from "model black boxes" and "model hallucinations." It also proposes exploring trusted evidence storage at critical transaction nodes. For consumer protection, the framework introduces more specific risk alerts and permission controls tailored to vulnerable groups, including the elderly and minors.
The release of this convention addresses a key innovation hotspot currently shaping the payment industry. Throughout this year, major payment and internet platforms have been highly active: Alipay launched its agent "Abao," WeChat Pay introduced an AI-exclusive card, and both JD.com and China UnionPay released their respective A2P2 protocol and APOP framework. These institutions are actively exploring diverse technical pathways in identity trust, authorization management, and capital segregation.
The rapid race for market share across the innovation landscape has made the clarification of rules increasingly urgent. Traditional payment risk control has relied on manual intervention or pre-set rules, yet the autonomy of intelligent agents breaks down the previous trust assumptions. The introduction of the KYA mechanism and pre-deployment reporting requirements effectively establishes both technical and compliance boundaries for AI payments ahead of large-scale industry commercialization.
According to available information, the convention was reviewed and approved at the second meeting of the Fifth Standing Council of the China Clearing and Payment Association. Going forward, under the guidance of the People's Bank of China, the association will organize supervision and inspection of member institutions' compliance. With these regulations now in place, agent-based payments are set to enter a new phase where compliance and innovation are pursued with equal emphasis.