Ant International has open-sourced a protocol intended to let AI agents operate within mobile-payment systems, moving the company beyond the familiar role of a wallet provider toward the harder task of defining how software agents can be identified, authorized, and constrained when money is involved.
The project is called the Agentic Mobile Protocol, or AMP. SCMP reported that its initial rollout connects with 10 Alipay+ digital wallets serving 1.5 billion accounts. The listed networks include Alipay, AlipayHK, DANA, GCash, KakaoPay, Toss, MPay, TNG eWallet, TrueMoney, and Starryblu.
That figure should not be read as 1.5 billion people suddenly using autonomous agents for payments. The networks provide the potential distribution footprint; implementation within each wallet remains an early-stage question. The news is important because Ant International has put a common framework into the open at a time when AI agents are beginning to move from answering questions toward carrying out defined tasks.
AMP links agent behavior to payment permissions
Ant International has published AMP under the Apache-2.0 license through its GitHub repository. Open-sourcing the protocol gives developers, payment operators, and outside reviewers a concrete starting point for examining how the system handles instructions and permissions. It also signals that Ant International wants the framework to become interoperable infrastructure rather than a feature limited to one proprietary app.
The central problem is authorization. A user may want an AI agent to look for an item, compare offers, and take a limited action, but that does not mean the agent should have unlimited access to the user’s funds or be able to make unrelated purchases. AMP describes a “Mandate Chain” that uses selective-disclosure JSON Web Tokens to carry task authorization. In practical terms, the protocol is designed to let an agent show what it is permitted to do without exposing more user information than the task requires.
This is different from adding a payment button to a chatbot. A payment button still leaves the user at the moment of transaction. An agentic-payment system has to define what the agent can do before, during, and after a request. It must make the relationship between the user, the agent, the merchant, and the payment network legible enough for each party to accept the transaction.
That is why Ant International’s work with payment incumbents matters. On September 10, the company, Visa, and Mastercard announced a “Know-Your-Agent” interoperability effort. Ant International describes the initiative as work toward common identity and verification standards for agents. The announcement does not establish a finished global rulebook, but it recognizes that agent identity cannot be solved by a wallet operator acting alone.
Alipay+ gives the protocol a regional test bed
Alipay+ gives AMP a potentially useful starting environment because it already links multiple Asian wallets. Ant International is not proposing to build a new consumer payment network from scratch. It is trying to add an agent-permission layer to a cross-border system that has existing wallet brands, merchants, and payment flows.
The initial 10-wallet list is therefore a more relevant indicator than a generic claim about AI commerce. The networks differ by market, user base, and operating context. If AMP is adopted, it will need to work across those differences while preserving the rules that make payment systems dependable. The protocol’s value will depend on whether a task authorized through one wallet can be understood by a merchant or service operating across the wider network.
China’s technology companies have already been racing to embed agents into work software and consumer services. EastFrontier’s report on AI agents entering enterprise workflows described how companies were turning model capability into task-specific products. AMP shifts the emphasis from an agent’s ability to reason over information to the conditions under which it can take a transactional action.
The difference is material. An enterprise agent may be useful if it drafts a report or organizes internal data. A payment agent has to interact with money, consumer consent, merchant rules, and fraud controls. The operational standard is therefore higher, even when the task appears simple.
The terms “open source” and “initial rollout” should be kept separate. Ant International has made AMP available for developers to inspect and use under an open license. SCMP reported the connection to 10 Alipay+ wallets. Neither fact confirms that every wallet user has access to agentic shopping, that every merchant can accept an agent-initiated payment, or that regulators across all affected markets have approved a common implementation.
Those gaps are normal at this stage. Payment protocols require adoption by different actors with different incentives. Wallet operators need to decide how much control to expose. Merchants need to recognize authorized agents. Users need clear ways to set limits and revoke permissions. Regulators need to understand the accountability chain when an automated action is disputed.
The Know-Your-Agent work with Visa and Mastercard addresses part of that challenge by focusing on common identity and verification. But it is still an interoperability initiative, not evidence that a universal standard has been implemented. Ant International’s next test will be whether it can translate the AMP specification into a product experience that is understandable to ordinary wallet users.
EastFrontier’s coverage of DeepSeek’s harness for autonomous AI agents examined the software layer that allows agents to carry out structured work. AMP adds a payments layer to the broader agent stack. One effort concerns how agents can be built and directed; the other concerns whether those agents can be trusted to move through a financial transaction.
Why China’s AI commerce push is taking a protocol route
Ant International’s strategy is notable because it frames agentic commerce as a standards question. The company is not only presenting a consumer feature. It is offering a shared protocol, an open repository, and a cross-network deployment path. That approach could make sense in a regional payment environment where no single wallet controls every user, merchant, or market.
The initiative also gives China-origin fintech infrastructure a role in the emerging discussion around AI agents and commerce. The competition will not be decided only by which model writes the best recommendation. It will also depend on which systems can demonstrate consent, limits, identity, and auditability once an agent goes from advising a consumer to acting for one.
Ant International has supplied the first pieces: an open-source protocol, a link to 10 Alipay+ wallets, and an identity effort with Visa and Mastercard. The remaining work is substantial. Developers and payment partners must turn the specification into reliable flows, while users and regulators must be able to see exactly what an agent is allowed to do. AMP is a meaningful infrastructure move, but its real value will be measured in the safeguards and adoption that follow.