HSBC and Ant Digital Successfully Test AI-Driven Tokenized Deposit Payments

HSBC and Ant Digital Technologies just ran a live test of AI agents making micropayments through tokenized bank deposits on a blockchain testnet. Not a concept paper. An actual trial.

The mechanics are worth unpacking. HSBC’s Tokenised Deposit Service handled the settlement side, processing transactions in real-time on what’s called the Jovay Testnet — a blockchain testing environment, not a live production network. Ant Digital brought its Anvita Flow network into the mix, which is basically a layer that lets AI agents discover available digital services and interact with them autonomously. During the trial, an AI agent picked a digital service, paid for it, and the whole thing settled without a human touching the transaction. The payment qualified as a micropayment — amounts typically under $2 — but the point wasn’t the dollar value. It was the plumbing.

HSBC also provided real-time risk assessment capabilities during the test. So the system wasn’t just moving money blind; it was running checks on the fly.

Banks Are Lining Up to Test This

HSBC and Ant Digital aren’t alone. Back in March, Santander ran a controlled test with Mastercard’s Agent Pay, and that one actually used live systems — an AI-agent-initiated payment that went through successfully. Swiss digital asset bank Sygnum has explored similar ground, and so has Spanish banking group CaixaBank. Both Sygnum and CaixaBank built in a customer approval step for each transaction, which is a different design philosophy — more cautious, more human-in-the-loop. The HSBC trial didn’t require that kind of manual checkpoint, at least not in what’s been shared publicly.

The fact that multiple major institutions are running these tests within the same window isn’t a coincidence. Banks have watched AI agent frameworks mature quickly over the past couple of years, and the question of how money moves in an AI-driven economy is becoming urgent. Traditional payment rails weren’t built for machines operating at machine speed. They were built for humans approving things during business hours. That mismatch is exactly what’s driving the experimentation.

Not everyone thinks legacy banks can actually make the leap.

Augustus Bank CEO Ferdinand Dabitz put it bluntly — many banks’ infrastructures were designed for human-operated processes, and automated transactions may not run optimally on those systems. He’s probably right that retrofitting decades-old core banking infrastructure for always-on AI agents is harder than it sounds. Whether that skepticism applies to HSBC specifically is unclear, but it’s a fair structural critique of the broader industry.

See also: IMF Warns of Major Interoperability Issues as Tokenized Finance Reaches $65 Billion

Citrini Research’s “Breaking The Wall” Report

On October 8, investment research firm Citrini Research dropped a report called “Breaking The Wall” that digs into exactly this territory. Their argument: as AI agents take on more transactional responsibility, demand for programmable financial infrastructure will climb. Blockchain networks, per Citrini, could supply the continuous, programmable foundation that AI-driven commerce needs. It’s not a radical idea, but it’s one that’s gaining traction in places where serious money is being allocated.

The timing of the report — landing the day before the HSBC-Ant Digital news broke — is kind of striking. The research wasn’t commissioned by either company, as far as the source material goes, but it frames the same problem both teams are trying to solve. AI agents need financial rails that don’t sleep, don’t require human sign-off on every action, and can handle high-frequency, low-value transactions without friction.

Tokenized deposits are one answer to that. They’re bank-issued digital representations of fiat money that settle on a blockchain, which means they can move programmatically. Combine that with a network layer that AI agents can query to find services — which is what Anvita Flow does — and you’ve got something that at least resembles a working prototype for machine-to-machine commerce.

Still early days, though. The HSBC and Ant Digital trial is explicitly a technical verification. No commercial rollout. No live customers. The Jovay Testnet is a controlled environment, and what works there doesn’t automatically translate to production at scale. Risk management, regulatory compliance, fraud detection — all of that gets harder when the volume and speed increase by orders of magnitude.

What’s notable is that the trial happened at all. HSBC is one of the largest banks on the planet. Ant Digital sits inside one of the most sophisticated fintech ecosystems in Asia. When those two run a joint blockchain test focused on AI agent payments, it’s not a side project. It’s a signal about where resources are going.

The micropayment amount — under $2 — might seem trivial. But micropayments at machine scale, running continuously, add up fast. And the infrastructure question isn’t really about any single transaction. It’s about whether the financial system can handle a world where AI agents are economic actors in their own right.

That question doesn’t have a clean answer yet. What HSBC and Ant Digital have is a testnet result.