JD.com has released China's first agent autonomous payment protocol — A2P2 (Agent Autonomous Payment Protocol). This isn't another "AI + payments" concept. It defines agent identity authentication, payment authorization, and transaction confirmation, with an ARI (Agent Runtime Identity) mechanism.
Six Levels of Autonomous Payment
| Level | Autonomy | Example |
|---|---|---|
| L0 | None | Human does everything |
| L1 | Info retrieval | AI finds products, human pays |
| L2 | Initiate | AI starts payment, human confirms |
| L3 | Rule-bound | Autonomous within preset budget |
| L4 | Conditional | Complex decisions with post-hoc reporting |
| L5 | Full | AI decides and executes everything |
L3 means you tell your Agent "buy groceries under 500 yuan" and it handles everything. L4 handles complex scenarios like "clear inventory above cost price."
ARI: Digital ID Cards for Agents
Each Agent gets a unique runtime identity — traceable, revocable, and auditable. Every transaction links to a specific Agent instance.
Why This Matters
Existing payment systems require biometric authentication — faces, fingerprints, SMS codes. Agents don't have these. They need entirely new payment infrastructure.
JD.com's A2P2 provides the first version. Given JD's e-commerce ecosystem and payment system, this protocol has real soil to grow in.
Risks
- Security: What if an Agent is attacked?
- Liability: Who pays for wrong orders?
- Regulation: Financial laws don't cover "non-human payment entities"
The tiered design addresses these — L3+ requires strict preset rules and post-hoc reporting. But edge cases will be the real test.
Auto-compiled by Deskless Daily. Full bilingual version at wdsega.github.io
Top comments (0)