Anne Boden founded Starling Bank in 2014 and built it into a leading British digital lender before stepping down as chief executive in 2023. She left the board the following year, and last month she sold the last of her nearly 5% stake, ending her ownership link to the bank she created. The Sunday Times estimates the proceeds of her Starling share sales at about £170 million, and a large slice of that money is now going into her second act.
That second act targets a gap Britain's banks have declined to close. AI assistants can already draft emails, compare prices and assemble shopping baskets, but they cannot reach a customer's current account to pay for anything. Large UK banks have refused to give agents that access, fearing unauthorised transactions and the liability questions that follow. The distance between what an assistant can decide and what it can actually settle has stayed wide.
The Sunday Times reported on November 15, 2026, in a piece headlined "Why I'm building my second bank in a decade", that Boden's answer is Lockwood. The company behind it, Lockwood AI Limited, is registered in England and Wales under company number 16500948, with a registered office in High Wycombe, Buckinghamshire. Its website is precise about status: Lockwood is not open yet, joining the waitlist does not open an account, it is not a bank, and it is not connected with Starling.
The product it describes is narrow by design. Lockwood will offer a UK current account for individuals and small businesses, with no lending and no savings products. It will not charge customers, and it intends to earn money from interest on deposits instead. Boden's plan is to start with an electronic money institution licence, which is quicker to obtain than full banking authorisation, and to convert Lockwood into a bank later.
Key Facts
The mechanics rest on permissions rather than products. Customers choose which AI assistants they trust and hand them routine money jobs: paying bills, cancelling forgotten subscriptions, sending birthday flowers. Each assistant receives a defined purpose and a spending ceiling. Payments inside those limits proceed automatically, while anything larger waits for a tap from the customer. Lockwood's own AI watches the account day and night, checks that every assistant is who it says it is, and steps in when something looks wrong. Boden's formulation is that Lockwood is a bank "built by agents, to serve customers that want to use agents, and then be defended by agents".
Boden's worked example involves Meta's planned UK shopping assistant, Muse. A customer might allow Muse to buy groceries but cap it at £100 a month. If Muse then presents a £200 fuel transaction, the payment is refused, because it falls outside the instruction. That example carries the whole proposition: the limit is not a balance check but a statement of intent, and the bank enforces the intent rather than merely the number.
Payment Expert reported on November 16, 2026 that the venture is funded personally for now, with outside capital to be raised next year, and that the operation runs on a team of 36, including "some consultants, very few employees". Even the name came from the technology. A team of AI agents negotiated with each other to produce a shortlist and settled on Lockwood because, in Boden's account, it is easy to say, easy to spell, and contains the word lock, which she associates with safety.
FinTech Futures reported on November 16, 2026 that Boden will apply for an electronic money institution licence as soon as possible and convert to a full bank afterwards, with the licence hoped for in 2027 and a bank launch in 2028. She argues that a greenfield build is easier to defend than retrofitting controls onto an incumbent's estate, describing large banks as castles with 400 doors of different sizes bolted on over time, against her plan for one door built with new technology that has no gaps.
Dealroom News reported on November 14, 2026 that Boden is funding the venture herself from roughly £170 million of Starling share sale proceeds, that she intends to expand internationally sooner than Starling did because an international business is easier to float, and that an eventual IPO sits on her roadmap. Her customer target is one million in Lockwood's first 12 months, and she has said the company will be orders of magnitude bigger than Starling, which has 6.3 million customers.
Analysis
What this really means is that the scarce asset in agentic finance is not the current account but the delegated authority attached to it. Opening an account is a commoditised problem with thin margins. Encoding what an agent is allowed to want, and refusing transactions that violate that instruction even when the balance would cover them, is the part nobody has standardised. Boden is not really selling banking. She is selling a permission layer that happens to settle payments.
That reframing explains the odd shape of the business. A free account with no lending and no savings looks thin beside a universal bank, but it is a clean substrate for a control plane. Interest on deposits is a modest revenue line that depends on rates staying high enough to produce a spread, and it gives Lockwood little incentive to push credit or upsell products an agent might be talked into buying. The absence of lending also keeps the capital and conduct risk profile simple, which matters when the selling point is that machines cannot be persuaded into a mistake.
The competitive question is harder. Incumbent lenders, from Britain's high street names to global balance sheet giants such as JPMorgan Chase, already own the customer relationship and can bolt on agent controls once demand is proven. The castle with 400 doors is also a castle with 400 distribution channels. Lockwood's advantage has to be speed and a clean liability story, not copyable features. The bigger picture here is that the assistant layer, rather than the bank layer, may end up owning the customer interface, and a bank designed to sit underneath Meta's Muse or any rival assistant is accepting a subordinate position in exchange for volume.
Regulatory risk is the binding constraint. An electronic money institution can hold funds and execute payments but cannot lend, which fits the plan, yet the conduct questions are formidable. If an agent buys 20 pairs of socks instead of two, or renews insurance with inadequate cover, who is responsible? Boden's own essay on the Lockwood site, dated November 14, 2026, concedes that an agent may remain within its financial limit while acting completely outside the customer's intentions. That sentence is both the design brief and the first test a supervisor will apply. Starling's £29 million FCA fine in 2024, for failings in financial crime controls, is a reminder that regulators judge the plumbing rather than the pitch.
Why It Matters
The immediate stakes are consumer protection. Agentic payments move the moment of decision away from the cardholder, and existing chargeback and fraud rules assume a human pressed a button. If Lockwood can show that a narrow permission, verified and monitored by a second AI, produces fewer disputes than conventional card controls, it builds an argument supervisors and other banks can adopt. If it cannot, the experiment becomes a cautionary tale about automating intent.
A market structure argument follows. Small businesses are the obvious first customers, because routine payments, subscription sprawl and repeat purchases are exactly the tasks an assistant handles well, and because a free account with automated housekeeping cuts the time cost of manual administration. One million customers would give the UK a reference implementation for agent banking before other jurisdictions settle their own approaches, which is why the licence timeline matters beyond one startup's balance sheet.
Finally, the venture tests whether trust can be designed rather than inherited. Boden's argument is that a bank born after agents exists can be hardened against them in a way that an older institution cannot. The claim is falsifiable, and the FCA, the payment schemes and the assistant providers will all get to check it in public.
Next Up
The near term calendar is set by the regulator. Boden hopes to secure the electronic money institution licence in 2027 and to launch the bank in 2028, while raising outside capital next year and turning her personal funding into an institutional round. Lockwood has already opened a waiting list, which gives it a demand signal to take to investors before it holds a licence to operate.
Alongside the application, Boden wants to bring banks and regulators together in a working party to decide how agentic banking should work. There is momentum behind it: Mastercard announced a UK sandbox for testing AI payment agents in July, four months after Santander and Mastercard completed Europe's first AI agent payment on live banking rails. The standards those groups write, and the limits they decide an agent may never cross, will determine whether Lockwood's single door becomes the template or the exception.
Comments (0)
Log in or sign up to leave a comment.
No comments yet. Be the first to share your thoughts.