HSBC and Ant Digital Test AI agent micropayments on Tokenized Deposits

HSBC and Ant Digital Technologies announced on October 9 that they have successfully completed a technical verification test of AI agent micropayments using tokenised bank deposits. The joint test, detailed in a PR Newswire announcement carried across the Asia-Pacific region, combined HSBC's Tokenised Deposit Service with Ant Digital's Anvita Flow network and its Jovay Testnet blockchain environment to show an AI agent discovering a digital service, paying for it with tokenised deposits, and settling the transaction in real time. It is one of the most concrete demonstrations yet of AI agent micropayments on bank-owned rails.

According to the announcement, the test demonstrated an end-to-end process in which an AI agent could discover and use a digital service, make a micropayment — typically defined as less than $2.00 — and complete the transaction with real-time settlement. HSBC contributed its Tokenised Deposit Service and related settlement capabilities, including real-time risk checks delivered through its MCP interface. Ant Digital's Anvita Flow provided the infrastructure that let the agent discover and invoke services, while the Jovay Testnet served as the Layer 2 blockchain environment for the transactions. Together, the companies said, the components demonstrated a potential end-to-end model for AI agent micropayments using tokenised deposits — a blueprint other banks could adopt as AI agent micropayments move from pilots to products.

Inside the test: discovery, payment and settlement in one loop

The significance of the test is that it closes the loop for AI agent micropayments. Many agent-payments experiments to date have handled payment as a bolt-on step: the agent decides what to buy, and a separate rails layer moves the money. In the HSBC–Ant Digital design, the agent manages service interactions and the associated payment as one connected, automated process — finding the service, invoking it, and settling the micropayment without a human in the loop.

HSBC's contribution centres on its Tokenised Deposit Service, which converts conventional bank deposits into on-chain tokens that can move on a blockchain while remaining liabilities of the bank. That distinction matters: tokenised deposits keep the transaction inside regulated banking infrastructure rather than shifting it to a stablecoin or crypto wallet. The bank added real-time risk checks through its MCP interface, the same interface standard that AI agents increasingly use to reach tools and data. Ant Digital's side supplied the agent-facing layer — Anvita Flow, which enables agents to discover services and manage settlement, and the Jovay Testnet, a Layer 2 testing environment where the transactions actually cleared.

Both companies stressed that the exercise was a technical verification only. As reported in the announcement, it was carried out solely as an exploration of the technology and does not represent a commercial launch or a live customer proposition.

Why banks are racing to own the agent-payment rail

The HSBC–Ant Digital test caps a busy 2026 for bank-led AI agent payments infrastructure, and each step has pushed the decision further toward the agent. According to industry reporting, Santander ran an AI-agent-initiated payment via Mastercard's Agent Pay in March, Sygnum moved AI-agent transactions to a blockchain mainnet with per-transaction customer approval and signing in May, and CaixaBank completed an AI-agent-initiated card transaction through Visa Intelligent Commerce. The HSBC–Ant Digital test is the first in that sequence to pair tokenised deposits with autonomous service discovery — arguably the purest version of AI agent micropayments attempted by a global bank so far.

The commercial logic is visible in the volume numbers. Coverage of the test notes that research from Citrini Research, published October 8, argues that autonomous AI agents could drive demand for programmable financial infrastructure, since systems designed for humans will have to adapt to machine-speed transaction patterns. One recent industry finding cited in the coverage counted more than 1.28 million agent-initiated payments in a single day across just 145 addresses — a preview of what settlement rails will have to absorb if agents become routine purchasers.

The broader ecosystem is moving in the same direction. Agent commerce toolkits are giving AI agents direct access to card networks, and retail integrations are connecting any shopping agent to checkout systems. What the HSBC–Ant Digital test adds is the settlement layer for AI agent micropayments: bank-native tokenised deposits instead of a card swipe or a wallet top-up. That architecture could eventually let agents pay fractions of a cent for API calls, data lookups or compute — the kind of machine-scale AI agent micropayments that card networks, with their fixed per-transaction fees, handle poorly today.

The gap between risk checks and judgment

The test also exposes the question banks still have to answer before AI agent micropayments can go mainstream. HSBC's real-time risk checks can flag whether a payment looks suspicious — but, as one independent analyst covering the announcement observed, they cannot answer whether the instruction behind the payment was wise. An agent that can discover services itself, choose among them, and pay autonomously has made a judgment call before any risk check runs. Who certifies the agent's choice of service, and who absorbs the cost when an agent buys the wrong thing efficiently?

The industry's answers are still forming. Agent identity and risk intelligence products are emerging to score agent-driven transactions, and regulators in markets like the UK are extending scrutiny from AI models to autonomous agents themselves. For now, the guardrail in the HSBC–Ant Digital test is architectural, and regulators will likely expect more before live AI agent micropayments are approved: the agent operates inside a testnet, against pre-scoped services, with bank-controlled settlement and risk checks on every transaction.

The executives behind the test framed it in those terms. Zhuoqun Bian, President of Blockchain Business at Ant Digital Technologies, said the verification demonstrates the potential use of on-chain AI agents and that the company will continue exploring how autonomous agents can create tangible value in real commercial environments, according to the announcement. Lewis Sun, HSBC's Global Head of Digital Currencies, said the bank is exploring how continued innovation and collaboration could enhance the customer experience and reduce friction for clients while keeping appropriate controls and responsible development at the centre.

As the joint announcement puts it, bringing AI, blockchain and regulated banking infrastructure together showed how the speed of software and the controls required by financial institutions could operate within a single transaction flow. A syndicated version of the announcement adds that the test validated the infrastructure needed to connect AI-driven decision-making, service access, payments and financial controls within one automated process. Whether that process — and AI agent micropayments more broadly — graduates from testnet to production will depend on the harder problems — agent identity, liability for autonomous purchases, and the economics of sub-$2 transactions at machine scale. But the direction of travel is unmistakable: the agent economy is getting its own payment rails, and banks want to build them.