JD.com has officially launched its Agent Autonomous Payment Protocol, referred to as the A2P2 protocol, marking the first such framework in China specifically designed for autonomous payments by AI agents.
The JD.com A2P2 protocol systematically categorizes autonomous payment capabilities into six levels, from L0 to L5. L0 represents payments requiring full human confirmation for every transaction, while L5 signifies fully autonomous payments by the AI agent. The protocol primarily focuses on the intermediate L3 and L4 levels. L3 allows an agent to autonomously initiate payment requests within a single task, with the system deciding whether to approve the request based on user-defined boundaries. L4 grants the agent greater authority, enabling it to complete payments autonomously as long as the amount, scenario, user, and other factors fall within pre-set parameters.
When an AI agent initiates a payment, fund security becomes a paramount user concern. The JD.com A2P2 protocol introduces an innovative ARI (Agent Runtime Identity) mechanism. This mechanism instantly binds three pieces of information at the moment of payment: the real user, the agent's identity, and the agent's runtime environment.
When an agent initiates a deduction request, the system immediately verifies three conditions: that the user ultimately bears responsibility for the payment; that the request is from the specific agent version uniquely authorized by the user; and that the agent is currently operating on a trusted device without malicious program injection. The request proceeds only if all three conditions are met.
Furthermore, the JD.com A2P2 protocol incorporates a segregated "fund carrier" layer. The user's main account is not directly accessible by the agent. Instead, a new, strictly limited "dedicated account" is created with hard rules governing maximum amounts, usage scenarios, validity periods, and permitted payees. Even if an agent is maliciously compromised, it cannot bypass these restrictions to access the user's primary account. Users can monitor how much funds are allocated to each agent and where the money is spent, and they can revoke authorizations at any time.
At the payment and settlement layer, the system immediately binds the payment result with the preceding task delegation credential, ARI identity, decision adjudication, and execution token, forming a complete evidential chain. To address the issue of critical payment data being scattered across different architectural layers—where traditional log analysis often provides incomplete information—the JD.com A2P2 protocol introduces a "deposit chain" as a unified factual anchor. This ensures every AI transaction is verifiable, manageable, and auditable.
A JD.com Technology representative stated, "Autonomous payment for AI agents requires ecosystem collaboration across the entire industry. We will advance the refinement and implementation of the JD.com A2P2 protocol in compliance with regulatory requirements. We will also partner with agent platforms, skill ecosystems, merchants, service providers, open-source communities, and technical developers to jointly develop supporting protocol standards, enabling seamless integration across different vendors and systems. Through these measures, JD.com will collaborate with partners to establish industry standards and explore application scenarios, ensuring AI technology genuinely serves the real economy."
Comments