AI is showing its receipts. Is your payments infrastructure ready?
Following on from my PAY360 wrap-up, one of my key takeaways from this year’s PAY360 was that "AI is finally showing its receipts".
It wasn’t on the formal agenda, but it was everywhere. In hallway conversations, panel discussions, and booth demos, the tone had shifted.
For the past three years, the industry has focused on what AI could do. This year, the conversation was grounded in what it is already doing: fraud detection, smart payment routing, reconciliation, customer experience. The hypothetical phase is over.
What comes next is not speculative either.
The next step is already being built
Agentic AI refers to systems that can pursue goals independently, make decisions and take actions across multiple steps with limited or no human intervention. Firms across financial services are already building and deploying these systems — using APIs to connect their AI agents directly into payment and account infrastructure.
In payments, that means client AI agents initiating, processing, and settling transactions end-to-end — interacting with payment infrastructure through APIs, without a human confirming each step. The intelligence sits with the client’s system. The infrastructure they depend on has to be unconditionally reliable.
As clients deploy agentic AI into their payment operations, it creates a fundamental shift in how they assess their banking provider. The question is no longer just about price or service. It is about whether the underlying infrastructure can be unconditionally relied upon when an AI system is in the loop — making the choice a risk decision, not just a procurement one.
The pressure point isn’t AI. It’s money under speed.
The problem is that most payment infrastructure was not designed for a world where autonomous systems initiate, process, and settle transactions at machine speed. The assumptions baked into legacy banking, that a human will review before funds move, that settlement can take a day, that reconciliation is a batch job, simply do not hold when AI agents are transacting continuously on behalf of clients.
The Mastercard AI panel at Pay360 framed it clearly: 'AI itself is no longer the interesting part. Managing the risk when AI operates in regulated payment flows is the real challenge'. The FCA's position is consistent with that view: firms cannot treat safeguarding as adequate just because nothing has gone wrong under calm conditions. The test is what happens when volume spikes, when an agent makes unexpected moves, when the system needs to demonstrate that every penny is exactly where it should be.
Ultimately when autonomous systems have access to significant pools of funds, the question shifts from:
Can the payment be made?
to
Can the funds be guaranteed to be there, every time, under any condition?
The question that isn't being asked loudly enough… Where do these funds actually sit?
Most non-bank financial institutions safeguard funds with clearing banks. Structurally, those funds support lending, which under normal conditions works but under stress can expose risks.
Some firms are exploring insurance as an additional layer. But insurance behaves differently to infrastructure. It is underwritten, repriced, and reassessed—often at the point where risk becomes most acute.
Infrastructure becomes the differentiator
If client AI systems are going to move money autonomously, the infrastructure beneath them has to guarantee two things simultaneously:
- Funds must be fully available on demand
- Movement of those funds must remain fully controlled
That combination is harder than it sounds: availability without control introduces risk and control without availability introduces failure.
Both have to operate at the same speed as the systems initiating the transactions.
Designing for a different reality
This is the context in which the Bank of London was built. Every pound deposited is held at the Bank of England, never leant, or leveraged. That means availability is structural, not conditional on a third-party balance sheet. As transaction velocity increases, that distinction matters. Continuous autonomous payment flows leave no room for uncertainty around access to funds. At the same time, control has to keep pace. Payment limits, permissions, and governance cannot sit outside the flow of transactions. They need to be embedded within it.
With direct access to UK clearing schemes and API-first infrastructure, the Bank of London is built to be the settlement layer that client systems connect to — not an operator of AI itself, but the infrastructure that AI-enabled clients can rely on. When a client’s system is making decisions at machine speed, the reliability of the underlying layer is not a secondary concern. It is the primary one.
Not a layer bolted onto legacy systems, but infrastructure purpose-built for a world where clients’ operating models have fundamentally changed — and where the settlement and liquidity of every pound matters more, not less, when the humans are no longer in the loop at each step.
The window is shorter than it looks
The changes to the Safeguarding regime introduced by the FCA and effective May 2026, creates a clear inflection point.
Daily reconciliation, monthly reporting, and annual audits are not administrative changes. They are designed to ensure firms can demonstrate - continuously, not retrospectively - that funds are exactly where they should be.
Under traditional conditions, weaknesses in infrastructure can remain hidden whereas under agentic conditions, they are exposed.
The combination of autonomous execution, continuous transaction flow, and real-time fund movement removes the tolerance for delay, ambiguity, or dependency.
This is where the divergence happens. Firms can build infrastructure aligned to this reality and meet the FCA’s expectaitons with confidence. Or they can approach it as a compliance exercise and discover, under pressure, that the underlying model was never designed for how their systems now behave.
What will matter next
The firms that lead in an agentic payments world will not necessarily be those with the most advanced AI.
They will be the ones that paired intelligent systems with infrastructure that can be relied on:
- when volume spikes
- when decisions are automated
- and when regulators ask for proof, not assurance
Because AI is accelerating how money moves. Infrastructure determines whether it holds.
