HSBC and Ant Digital Trial AI Agent Payments Using Tokenized Deposits
HSBC and Ant Digital have completed a technical trial in which AI agents autonomously discovered services and executed micropayments using tokenized deposits, with real-time settlement and risk checks running in parallel.
HSBC and Ant Digital have completed a technical trial in which AI agents autonomously discovered services and executed micropayments using tokenized deposits, with real-time settlement and risk checks running in parallel. The trial, announced Friday, marks one of the first publicly disclosed tests of AI-driven payment execution built on tokenized banking infrastructure at institutional scale.
Tokenized deposits are digital representations of commercial bank money held on a blockchain or distributed ledger, functionally equivalent to a standard deposit but programmable and transferable without the latency of traditional correspondent banking rails. In this trial, AI agents used HSBC's tokenized deposits as the payment mechanism, identifying services autonomously and completing transactions without step-by-step human authorization for each payment. Real-time settlement removes the typical T+1 or T+2 clearing window, while the embedded risk checks suggest the system can flag anomalous transactions before they settle rather than after.
The collaboration pairs two institutions with serious blockchain track records. HSBC has been building tokenization infrastructure since at least 2023, when it began trialing tokenized deposits for institutional clients and launched HSBC Orion, its platform for issuing digital bonds. Ant Digital, the blockchain and technology arm of Alibaba's fintech group, brings experience in high-volume digital payment infrastructure across Asian markets. Together, they are testing whether AI agents can function as credible financial counterparties rather than simple automation scripts executing pre-coded instructions.
Singapore's Project Ubin, the European Central Bank's digital euro pilots, and JPMorgan's JPM Coin have each demonstrated that tokenized money can cut settlement times and reduce operational costs in controlled environments. What distinguishes this trial is the AI layer on top. Previous tokenization experiments kept humans in the authorization loop; this one removes that step for micropayments, letting the agent discover, negotiate, and pay within a single automated flow. That is a meaningful architectural shift, not an incremental one.
The risks are proportionally significant. Regulatory frameworks in most jurisdictions do not yet clearly assign liability when an AI agent executes a transaction that results in a loss. Who is responsible: the bank that issued the tokenized deposit, the AI model developer, or the enterprise deploying the agent? That question has no clean answer under current law in the UK, EU, or Hong Kong, where HSBC and Ant Digital primarily operate. Security exposure is another live concern: an AI agent with payment credentials is a high-value target, and the attack surface expands whenever autonomous systems hold or transfer real monetary value. Scalability is also unproven. A single technical trial, however well-designed, cannot replicate the transaction volumes, edge cases, and adversarial conditions of a live payment network.
Interoperability presents a longer-term structural challenge. HSBC's tokenized deposit infrastructure runs on its own rails. For AI-agent payments to become a general-purpose capability rather than a bilateral experiment, those rails need to connect to other banks' tokenized assets, to central bank digital currency (CBDC) systems under development, and eventually to retail payment networks. That requires standardization work across institutions and regulators that no single trial can resolve.
Institutional appetite for AI-native payment infrastructure is real, and the combination of programmable money with autonomous agents points toward a future where machine-to-machine commerce operates at a speed and granularity that human-authorized payments cannot match. Subscription micropayments, API-call billing, real-time supply chain settlement: these are use cases that become tractable when an AI agent can pay for a service in the same moment it consumes it.
The immediate next steps will be telling. If HSBC and Ant Digital move toward a limited live pilot with actual customer funds, the regulatory conversations with the UK Financial Conduct Authority and Hong Kong's Monetary Authority will become the critical variable. Both regulators have signaled openness to innovation in digital money, but neither has published clear guidance on AI agent authorization in payment systems. How quickly that guidance arrives may determine whether this trial becomes a product or remains a proof of concept.
