AI cuts open banking build costs. Trust is another matter
OBL's AI consultation says AI could cut the cost of building open banking and proposes a certification model. Banks will still need to know who stands behind each agent.
Published:
min

OBL's AI consultation says AI could cut the cost of building open banking and proposes a certification model. Banks will still need to know who stands behind each agent.
Published:
min

Open Banking Limited (OBL) has published the findings from its AI Green Paper consultation, under the title "Do LLMs Have Standards?" The question behind it is a good one. UK open banking took a regulatory mandate on the nine largest banks and building societies and years of bespoke development to reach scale. Does the next phase, and the move into open finance, have to be built the same way?
OBL's answer is a careful no. Its findings also show where AI changes the work of trust in open banking and where it leaves that work exactly where it was.
OBL uses an efficiency gain of up to 30% as a planning assumption for well-governed, repeatable work such as data mapping, documentation and testing, and is clear that this is a hypothesis still to be tested. It now treats AI as one of the main customers of the Open Banking Standard and is moving the standard onto a single, machine-readable source that AI tools can connect to directly.
It also notes that the time and cost of building software products have fallen, so founders can take an idea to market with smaller teams in less time. The regulatory costs of authorisation and reporting stay where they are. The likely result is more providers, built faster, by smaller teams, many of whom will have written much of their code with AI.
Where agents act for customers, OBL says consent "should not be treated as a static, one-off permission." It should define "what an agent may do, for what purpose, against which account or service, over what period, subject to which limits, and with what revocation or escalation rights." The paper also asks openly "whether existing legislation is sufficient at assigning where responsibility sits if an agent acts incorrectly."
In his foreword, OBL chief executive Henk Van Hulle puts the overall point simply: "AI does not lessen the importance of high-quality data, security, accountability or trust. It makes them more important."
OBL sets out a proportionate, risk-based certification model with three layers. The first tests AI-ready standards documents. The second tests datasets that AI has helped prepare, including reconciliation to source and audit trails. The third tests the tools agents use, checking that they stay within delegated authority and can be stopped, revoked and evidenced.
OBL sees the Future Entity convening this trust model, setting minimum expectations and benchmark tests, "without becoming the sole certifier of every AI product or dataset." Shared minimum standards, with assurance coming from more than one place, is a sensible way to build it.
Each of those layers tests a thing: a document, a dataset, a tool. A bank deciding whether to act on an instruction from an agent will also want to know about the company that operates it. Who is it, is it authorised, is it still in good standing six months after it passed its checks, and who answers if its systems are breached?
Those answers change over time, so they need checking continuously and sharing with every financial institution and intermediary connected to that company. Network accreditation of this kind sits alongside FCA authorisation and OBL's conformance work. It adds the up-to-date view of the company that neither is designed to give.
Recent events in the US show what happens when banks can't get that view. Citi, PNC and Citizens have joined other large banks in making customers responsible for what outside AI agents do with their accounts, while Citi covers agents it provides itself.
OBL is handling consumer-facing AI agents in a separate workstream, and the FCA's Mills Review has recommended developing trusted agent standards within open finance. The groundwork in this paper, from consent with limits and revocation to audit trails and certified agent tools, is what that work will build on. Adding the company behind each agent to the picture now will be far easier than adding it after the first serious dispute.
That picture of the company, kept up to date, is what Invela provides. OBL's model is designed to confirm that a document, a dataset or a tool meets the standard when it is tested. A bank also needs to know if the company running that tool has since changed ownership, suffered a security incident or seen its financial position weaken. Invela assesses the companies in the network against a common standard and then keeps monitoring them in real time, so a change in a company's risk shows up when it happens instead of at the next review.
Every financial institution and intermediary connected to that company sees the same view. A provider assessed once can be relied on across the network, so institutions don't each have to repeat their own review before it can connect. As AI brings more providers into open banking, more quickly, that saves banks from re-running due diligence on every newcomer and saves providers from answering the same questions bank by bank.
It also matters when something goes wrong. OBL asks whether the law is clear enough on where responsibility sits when an agent acts incorrectly. Whatever the answer, settling it depends on knowing which company stood behind the agent and what its risk standing was at the time. Invela keeps that record, which is what lets liability land with the party responsible. It is also what would let a bank accept agents run by companies it didn't build, instead of leaving the risk with the customer as Citi has.
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.
Insight on accreditation, risk scoring and warranty for banks, intermediaries and third parties.
Your email is used only to send Invela blog updates. Privacy Policy