The trust inversion: what no single firm can fix alone

In an essay for the Kyndryl Institute, Colin Payne, Head of Innovation at the Financial Conduct Authority (FCA) and Chair of the Global Financial Innovation Network, names something many in financial services have sensed. He calls it the “trust inversion”: the infrastructure that lets AI agents act for customers, from identity to payment authority to the ability to start a transaction, has arrived before the agents themselves have earned the trust to use it.

The risk isn't the brilliant agent

A year ago, the worry was agents that were too capable. Payne argues the risk is more mundane: agents that aren't yet dependable. On one demanding independent benchmark of real office work, the strongest agent completes around a third of tasks on its own, and Gartner expects more than two in five agentic AI projects to be cancelled by the end of 2027. The danger, as he puts it, is “the mediocre agent, deployed too widely, trusted too soon.” And an agent's mistakes rarely stay local. They flow into every system and every other agent that relies on its output.

We need an accountability stack

Payne's answer is to build accountability in while agentic finance is still in pilots, rather than after the first scandals hit the front pages. He sets out what that means: a verifiable mandate traceable to a named principal, permissions with hard limits, a tamper-evident record of what was done, consent that can be withdrawn instantly, liability mapped in advance, a route to redress for customers, and a kill switch.

Most of that list belongs to individual firms and agent providers. Each can build its own audit trail, its own permission limits and its own way of switching an agent off. But two parts of the stack can't be built by any one firm on its own.

Liability can only be mapped in advance if everyone agrees the map

Payne frames the question succinctly: who pays when this fails? Answering it in advance means agreeing it across everyone in the chain, from the bank to the intermediary to the third-party provider to the company behind the agent. Today, outside a few schemes, open finance settles liability one bilateral contract at a time, and the agents now entering the chain never signed any of them.

Correlated risk is invisible from inside one firm

Payne's systemic concern is correlation. If thousands of firms build on a handful of models and cloud providers, they don't hold independent risks; they hold the same risk many times over. He notes that the Bank of England has begun simulating how agents might herd under stress, and that Parliament's Treasury Committee has recommended designating the largest AI and cloud providers as critical third parties. By definition, that kind of risk doesn't show up when a firm assesses its own suppliers. It only shows up when you can see across the network: who depends on whom, and how that is changing.

Agents won't respect borders

Payne also argues that cross-border coordination is infrastructure rather than a nicety, because an agent authorised in one jurisdiction can transact in another within the same second.

Liability agreed in advance, risk that only shows up across the network, and accountability that holds across borders have one thing in common: they sit between firms, not inside individual firms. A bank can build a kill switch for the agents it uses. On its own, it can't see that many of its partners depend on the same model provider, or agree in advance with every intermediary and agent provider who pays when something breaks.

That shared layer is what Invela is building: accrediting the companies in the open finance chain, including those behind AI agents, monitoring how their risk changes and where it concentrates, and giving everyone in the network the same evidence to settle liability before something fails, rather than argue over it afterwards.

We'll be publishing two reports in the coming weeks that explore these questions in more depth. One draws on our Powerhouse Debate at Open Banking Expo with UKPI, NatWest and the FCA, and looks at the liability and visibility gap in UK open finance. The other draws on our Money20/20 USA panel with MX, Moore & Van Allen and Spire, and looks at why US open finance needs network rules rather than one contract at a time. Sign up to our blog to get the links first.

Invela is the infrastructure layer that makes open finance trustworthy - accrediting who's in the network, monitoring risk in real time, and ensuring liability lands in the right place.