Open Banking Is Becoming the Rail for AI Assistants: What Read-Only Account Access Means for Banks, Fintechs and Brands
For most of its life, open banking has been infrastructure that customers never noticed. It powered budgeting apps, affordability checks and pay-by-bank buttons quietly in the background. That is changing. General-purpose AI assistants are now using open banking-style account aggregation to read customers' bank and card data, and that makes the rail suddenly strategic. In November 2026, Plaid announced it was powering Perplexity's "Personal CFO" in 17 European countries, and said it also supplies the bank-account link for personal finance features in Claude and ChatGPT. Anthropic launched a personal finance tool for Claude that reads linked accounts through Plaid in the US, according to Finextra. The pattern is clear: assistants are becoming a front door to customers' financial data, and read-only aggregation is the way in. This analysis explains how that access works, what it means for banks, fintechs and brands, and why the step from reading to paying is the contested one.
What has actually happened
Three moves in a short period show the direction.
Assistants are adding personal finance features. Anthropic's tool for Claude reads linked bank and card accounts through Plaid, in a US beta, according to Finextra's summary of Anthropic's announcement. Plaid says it supplies the bank link for personal finance in Claude and ChatGPT.
Aggregation has reached Europe. Plaid's own blog says it powers Perplexity's Personal CFO in 17 European countries, under a regulated account-information route. UK availability was not confirmed in that announcement.
Assistant owners are building managed channels for businesses. OpenAI introduced "Businesses in ChatGPT", with claimed business profiles, product feeds and business agents, in a US-first beta. That is about commerce rather than banking, but it shows assistant owners turning their interfaces into channels that businesses must manage.
The figures and claims in these announcements are the companies' own. Taken together, they show account aggregation becoming a common rail between assistants and customers' money.
How read-only access works
Read-only account access through open banking rests on three ideas that UK firms already know well.
Consent. The customer explicitly authorises a provider to access specified account data for a specified purpose. In the UK and Europe, that consent sits within a regulatory framework for account information services.
A regulated intermediary. An authorised provider, often an aggregator, connects to the customer's bank through APIs and passes the data on. The assistant does not hold the customer's banking credentials.
Scope and revocability. Access is limited to what the customer consented to, and the customer can withdraw it.
That is a very different model from the one banks have been warning customers against. Bank of America, for example, has told customers not to give banking credentials to AI assistants, according to TechCrunch. Credential sharing gives an assistant everything with no boundary. Consented aggregation gives it read access to defined data, through a regulated route that can be revoked.
Why "read-only" is the important word
Read access is a far smaller risk than payment access. An assistant that can see transactions can summarise spending, spot subscriptions or flag unusual charges. It cannot move money. That is why read-only access is spreading quickly while transactional access remains contested.
What it means for banks
For a UK bank, assistants reading accounts through aggregators is both a threat and a signal.
The threat is disintermediation of the relationship. If customers increasingly ask an assistant about their money rather than opening their bank's app, the bank loses the moment of engagement where it explains, advises and cross-sells. The data still comes from the bank; the conversation happens elsewhere.
The signal is demand. Customers want to ask questions about their money in plain language. Banks that build that capability into their own channels keep the conversation. Starling's "Talk to Transact", which lets customers initiate payments by voice and confirm with a tap, according to FF News, is one example of a bank building agent-style features inside its own app.
Practically, banks should expect a growing share of account-information requests to originate from assistant-led use cases through aggregators. That has implications for API capacity, consent dashboards, fraud monitoring and how clearly customers can see which services have access to their data.
What it means for fintechs
Fintechs built on account data face sharper competition and new partnership options at the same time.
Competition. A general-purpose assistant that can read accounts can offer basic budgeting and spending insights for free, as one feature among many. Fintechs whose main proposition is "see all your accounts in one place" will find that harder to defend.
Partnership. Fintechs with genuinely specialised capability, in lending, savings, insurance or financial planning, may find assistants become a distribution channel. The question is whether and how a regulated product can be offered inside an interface someone else controls, which raises its own authorisation, branding and perimeter questions.
Differentiation. The defensible ground is what an assistant cannot easily do: regulated products, specialist decisioning and depth in a particular customer need.
What it means for brands
For brands that are not banks, retailers, marketplaces, travel and loyalty businesses, the shift matters in two ways.
First, it confirms that customers are comfortable letting software see their financial lives when there is a clear benefit and a trusted route. That supports brands building embedded finance propositions, such as wallets, cards, loyalty-linked accounts or instalment products, under our Embedded Finance & Digital Wallets agenda, where account data can improve onboarding, affordability and personalisation with the customer's consent.
Second, it raises the question of how a brand's own financial products will appear to assistants. If a customer asks an assistant to review their spending, a brand's wallet or card will either be visible and well described, or invisible.
The contested step: from reading to paying
The step that matters most commercially is from read to pay. That is where the industry is far less settled.
Gatekeepers are cautious. Bank of America and United Airlines have restricted agents from transacting, according to TechCrunch and PYMNTS reporting. A vendor-authored article reported that several UK banks block AI agents, although the named banks and figures are unconfirmed. Fraud vendors are launching tools to detect agent traffic.
At the same time, sanctioned routes are being designed. UK payments voices have floated commercial variable recurring payments, pre-authorised mandates with limits the customer sets in advance, as a way to bound what an agent can pay, according to Open Banking Expo's coverage of a recent panel. Card schemes and issuers are building scoped agent credentials. Banks are building their own confirm-before-pay features.
For anyone designing products under our Agentic Commerce & Payments agenda, the practical conclusion is that read access is arriving now through established consent models, while payment access will come through narrower, rule-bound routes that are still being agreed.
Consent design when the reader is an assistant
Open banking consent was designed with apps in mind: a customer connects a budgeting tool, sees what it can access and can switch it off. Assistants put more pressure on that model, and four design questions follow for anyone building consent journeys.
Is the purpose specific enough? An app usually does one thing with account data. A general-purpose assistant can do many things, from summarising spending to suggesting products. Consent that says "personal finance features" is much broader than consent for a single budgeting function. Customers, and the firms involved, benefit from purposes that are concrete.
Who is the customer dealing with? In an assistant-led flow there may be the assistant owner, an aggregator and the bank. Customers need to understand which party holds their consent, which one they would complain to, and who they revoke access through.
Can the customer see the assistant on the bank side? Bank consent dashboards should show clearly when access was granted for an assistant-led service, so that a customer reviewing their connections recognises it.
What happens to data after the conversation? Data read to answer one question may be retained, combined or used to personalise later answers. Clear retention and use terms matter more when the reader is a general-purpose assistant rather than a single-purpose app.
None of these questions is unique to AI, and the existing consent framework already addresses each in principle. The point is that assistants make the answers matter more, because the scope of what can be done with the data is wider.
What to do now
- Banks: review consent dashboards and API monitoring for assistant-originated access through aggregators, and decide where your own channels should offer conversational, agent-style features.
- Fintechs: be clear about what your product does that a general-purpose assistant cannot, and explore whether assistants could become a distribution channel for regulated products.
- Brands: design embedded finance products so that, with customer consent, they are legible to assistants, and plan for how customers will ask about them.
- Everyone: treat credential sharing as a risk to design out, and consented, scoped access as the route to design for.
Where Innovify fits
Innovify is a London-based technology partner with deep expertise in banking, financial services, fintech and insurance, helping mid to large businesses, including brands that are not financial companies, embed financial services into the products their customers already use. We design compliance in from day one. Through our Embedded Finance & Digital Wallets work, we help banks, fintechs and brands design consented data access, account-linked products and the step from read to pay, as an implementation partner across multiple fintech platforms.
FAQ
How do AI assistants access bank accounts?
Most use consented account aggregation: the customer authorises a regulated intermediary to read specified account data, which is passed to the assistant. In November 2026, Plaid said it supplies the bank link for personal finance features in Claude, ChatGPT and Perplexity. The assistant does not hold the customer's banking credentials.
Is it safe to let an AI assistant read my bank account?
Consented, read-only access through a regulated intermediary is far safer than giving an assistant your banking password, because access is limited to agreed data and can be revoked. Banks have warned against credential sharing. Customers should check which provider is involved and what data they are agreeing to share.
What does open banking have to do with AI agents?
Open banking provides consented, scoped and revocable access to account data, which makes it a natural rail for assistants that need to read financial information. It may also support payments, for example through pre-authorised mandates such as commercial variable recurring payments, though that is still being discussed in the UK.
Can AI assistants make payments from my bank account?
Read access is spreading, but payment access is contested. Some banks restrict agents from transacting, while sanctioned routes such as pre-authorised mandates, scoped agent cards and in-app confirm-before-pay features are being designed. Where it exists, payment access is narrower and rule-bound.
What should banks do about assistants reading customer accounts?
Banks should expect more assistant-originated access through aggregators, review consent dashboards and API monitoring, and decide where to build conversational features into their own channels. Keeping the customer conversation in the bank's own app is the best defence against disintermediation.
Conclusion
Open banking was built to give customers control over who sees their financial data. That design is now meeting a new kind of reader: the AI assistant. Read-only, consented access is arriving quickly, and it changes who holds the customer conversation. The harder step, from reading to paying, will come through narrower, rule-bound routes. Banks, fintechs and brands that plan for both stages now will be better placed than those that treat assistants as a passing feature.
Related reading
- Open Finance vs Open Banking: What the Difference Means for a Brand Building Payments and Data Products
- Putting a Financial Product Inside Someone Else's AI Assistant: Authorisation, Branding and the Regulatory Perimeter
- Why Banks and Marketplaces Are Blocking Outside AI Agents, and the Sanctioned Routes Left Open
- Open Banking to Agentic Commerce: What a Decade of UK API Infrastructure Investment Teaches Us











