The first operational announcement for Ant International’s Agentic Mobile Protocol (AMP) has a telling shape: ten wallets and seven acquirers, but no switch-on date. The list published on September 11 frames the project for 2026 in accelerated terms, yet the partners’ statements remain conditional. That gap should not be read as simple PR caution; it separates an agreement in principle from a working payment network.

The wallets are Alipay, AlipayHK, MPay in Macao, DANA in Indonesia, GCash in the Philippines, KakaoPay and Toss in South Korea, TNG eWallet in Malaysia, TrueMoney in Thailand and Starryblu in Singapore. The acquirers are Adyen, Allinpay, Checkout.com, Fiserv, Global Payments, Nuvei and Worldline. The announcement visual shows eight logos, with Antom, Ant International’s own acquiring business, as the eighth. Tech Wire Asia has asked Ant to confirm whether Antom is included in the total. That detail matters, because the boundary between platform and protocol participant remains ambiguous.

The partners’ commitment is narrower than the framing suggests. The wallets will support AMP within their own agent security architecture: identity and authorisation, not agentic checkout. TNG Digital’s Alan Ni talks about exploring, while TrueMoney’s Koravut Pavitpok sees an opportunity to study. Only Starryblu and MPay use language that hints at something already running. TNG Digital declined to comment on timelines, regulatory engagement or liability.

The wallet bet, not the card bet

Competing protocols run on cards. OpenAI and Stripe’s Agentic Commerce Protocol handles the checkout handshake, the Agent Payments Protocol (AP2) covers payment consent, and Visa’s Trusted Agent Protocol extends card rails to agents. Mastercard announced Verifiable Intent in March 2026, co-developed with Google, to create a tamper-resistant record of what a user authorised an agent to do. Checkout.com’s Brian Sze makes the point directly in Ant’s announcement: agentic commerce has so far been largely card-based, and AMP extends it to wallets. The absence of non-Asian wallets is not accidental. In markets where QR wallets carry the volume that cards carry elsewhere, a card-first agentic stack has nothing to plug into. Southeast Asia becomes the proving ground for structural reasons, not just commercial ones.

The scale claims need careful reading. Ant says the Phase I wallets serve 1.5 billion user accounts and that the wider Alipay+ ecosystem connects 150 million merchants to two billion consumer accounts. Those are accounts, not people; Alipay in China accounts for a large share. Since April the figures have shifted: Ant previously cited more than 40 wallet partners and 1.8 billion accounts, while the merchant count has not moved. That mix makes it hard to tell how much of a real network AMP already is.

Governance and liability are still the sticking point

On governance, the announcement adds work on a Know-Your-Agent interoperability framework with Mastercard and Visa, inside BuildFin.ai, the platform convened by the Monetary Authority of Singapore, and on the Safeguards for Agentic Finance at Runtime. The stated goal is agent onboarding and identification across networks, while each network keeps its own verification and decisioning. But there are no timelines, pilot volumes, participating merchants or technical specifications. Mastercard and Visa supplied no quotes. Ant was already present in the card camp: in April it said it was among the first partners in Mastercard and Visa pilots for card-based transaction capabilities for AI agents, and it was working with Google on agentic commerce protocols.

The real difference is control. Google handed AP2 to the FIDO Alliance on April 28, 2026, releasing version 0.2, and Mastercard contributed Verifiable Intent to the same body. The FIDO payments working group is chaired by Mastercard and Visa. AMP, by contrast, has source code, SDKs and documentation on GitHub, but the protocol remains under Ant International’s control; there is no commitment to move it to a neutral standards body.

Then comes liability, where ambiguity becomes concrete. AgentSafePay, the money-back mechanism attached to AMP transactions, is described in three different ways in Ant’s materials. In the September feature list it is a guarantee for merchants against agent-specific risks. In the backgrounder it covers all AMP-based transactions and protects users from additional risk. In April’s release it was a mechanism for payment partners in cases of account takeover. Merchants, users and payment partners face different exposures; the documentation does not settle who the guarantee answers to. In a fragmented regulatory landscape, that is not a minor issue. Singapore operates with BuildFin.ai and SAFR; Malaysia, Indonesia, Thailand and the Philippines have no specific framework for payments initiated by an AI agent on a user’s behalf. If an agent gets it wrong, it is still not clear who pays, and on what legal basis.

The stakes go beyond a slice of the $3 trillion to $5 trillion market cited by McKinsey. They include the possibility of building an authorisation layer that does not start from cards and is not governed by card networks. That is why the silence on dates is the real story: until a wallet switches the capability on, AMP remains a set of exploration agreements. When one does, it will be the first observable test not only for Ant, but for an ecosystem trying to learn whether an alternative to card rails is credible.