Ant International Open-Sources Protocol to Let AI Agents Pay Through Digital Wallets

A Technical Play for the Autonomous Commerce Layer

Ant International released its Agentic Mobile Protocol to the public on 11 September, placing the specification on GitHub as an open standard. The protocol is designed to handle a specific technical challenge: how an autonomous AI agent can securely authenticate and execute a payment through a mobile wallet without requiring the user to manually approve each transaction. At launch, the protocol supports ten digital wallet platforms operated by Ant International, including AlipayHK, Starryblu, KakaoPay, Toss, and six others across Asia-Pacific markets.

The move positions Ant's overseas division as an early mover in what the company expects will become infrastructure for agentic commerce, where AI assistants negotiate purchases, compare prices, and complete transactions on behalf of users. At Opentechwire, we've tracked similar infrastructure bets across the region, from Singapore's PayNow interoperability experiments to Seoul's push for digital-identity-linked payments. Ant's release stands out not for novelty in concept but for the breadth of its wallet network and the timing: it arrives as foundation-model providers race to ship agents capable of multi-step tasks that involve real-world consequences.

How the Protocol Handles Authentication

The Agentic Mobile Protocol sits between an AI agent and a mobile wallet's authentication layer. When an agent decides to purchase an item, AMP generates a secure request that the wallet app can parse and verify. The user receives a notification and can approve or reject the transaction through biometric authentication or a PIN, depending on the wallet's configuration. Once approved, the agent receives a tokenised confirmation and proceeds with the purchase.

The protocol does not replace existing payment rails. It operates as an abstraction layer that standardises how agents request permission and how wallets respond, regardless of whether the underlying payment flows through a card network, bank transfer, or closed-loop system. This design allows Ant to extend AMP across wallets that use different backend infrastructure without requiring each to rebuild their authorisation logic.

Ant International built the protocol on top of its existing cross-border payment infrastructure, which already connects its portfolio of wallets to merchants in over 60 markets. The open-source release includes reference implementations in Python and JavaScript, along with API documentation that third-party developers can use to integrate AMP into their own agent frameworks.

The Wallet Network and Regional Footprint

The ten wallets that support AMP at launch represent a mix of Ant International's direct investments and strategic partnerships. AlipayHK serves Hong Kong; KakaoPay and Toss operate in South Korea; Starryblu is active in Southeast Asian markets. The full list also includes wallets in Pakistan, Bangladesh, and the Philippines, covering a combined user base that Ant has previously disclosed exceeds 400 million accounts across its international portfolio.

This footprint gives Ant a structural advantage over protocol proposals from single-market players. A developer building an AI shopping assistant can integrate AMP once and gain access to payment authentication across multiple countries, rather than negotiating separate agreements with each wallet operator. The protocol's value increases with the number of participating wallets, and Ant's existing relationships lower the friction for expanding that network.

However, the protocol's reach remains limited to markets where Ant International has established wallet operations. It does not yet include integration with major Western payment platforms or wallets operated by competitors such as Grab, Gojek, or Paytm. Ant has not announced whether those providers have been approached or whether they would be willing to adopt a standard controlled by a competitor.

Competitive Context and the Standards Race

Ant International is not alone in attempting to define how AI agents interact with payment systems. Payment processors including Stripe and Adyen have begun piloting agent-oriented APIs, while Apple and Google have explored ways to let their respective voice assistants complete purchases through Apple Pay and Google Wallet. None of these efforts have yet converged on a shared standard, and the open-source release of AMP can be read as Ant's attempt to set the terms before a dominant protocol emerges.

The decision to open-source the protocol is tactically significant. By publishing the specification and reference code, Ant lowers the cost for other developers to adopt AMP and raises the cost for competitors to fragment the ecosystem with incompatible alternatives. If AMP gains traction among agent developers, wallets outside Ant's network may face pressure to support it in order to remain accessible to autonomous commerce applications.

At the same time, open-sourcing does not guarantee adoption. Developers will choose a protocol based on the wallets it supports, the quality of its documentation, and the trust they place in its governance. Ant's decision to host the project on GitHub suggests it intends to accept external contributions, but the company has not yet published a governance model or clarified how decisions about the protocol's evolution will be made.

Security and User Control Questions

The protocol introduces a new attack surface. If an AI agent is compromised or behaves unexpectedly, it could attempt to initiate payments without the user's intent. AMP's design requires explicit user approval for each transaction, which limits the risk of fully autonomous spending. However, this approval step also limits the protocol's usefulness for high-frequency, low-value transactions where manual confirmation would create friction.

Ant has not disclosed whether AMP will eventually support pre-authorisation models, where a user grants an agent permission to spend up to a certain limit within defined parameters. Such models are common in subscription and recurring-payment contexts, but they introduce additional risk when applied to autonomous agents whose decision-making logic may be opaque or difficult for users to audit.

The protocol's security also depends on the robustness of the underlying wallet apps. A vulnerability in a wallet's authentication flow could be exploited by a malicious agent or a compromised device. Ant International has not published a security audit of AMP, and the open-source release does not yet include formal verification of the protocol's cryptographic components.

What This Means for Agentic Commerce

Ant's move reflects a broader shift in how payment companies are positioning themselves relative to AI. Rather than waiting for agents to become widespread and then adapting, Ant is building the infrastructure in advance and offering it as a public standard. If the protocol succeeds, Ant's wallet network becomes a default channel for autonomous commerce in the markets it serves. If it fails to gain adoption, the cost is limited to the engineering effort and the reputational risk of promoting a standard that others ignore.

The release also signals Ant International's strategy for differentiation in a crowded digital-wallet market. As wallets in Asia-Pacific mature and user growth slows, operators are searching for new sources of transaction volume. Enabling AI agents to use their platforms positions these wallets as infrastructure for the next generation of commerce applications, rather than simply mobile versions of credit cards.

Whether developers and competing wallet operators will adopt AMP remains uncertain. The protocol's technical merits will be tested in the months ahead as agent frameworks evolve and as users begin to trust autonomous systems with financial decisions. For now, Ant has made a public bet that the infrastructure for agentic payments is worth standardising early, and that its wallet network gives it the leverage to define how that standard should work.