A UK businessman lost £14,244 from his Metro Bank account after fraudsters used his debit card to buy credits for Anthropic’s Claude AI chatbot — even after he told the bank the transactions were not his. The case, reported by The Guardian on 4 August 2026, exposes a gap between banks’ fraud-detection promises and what actually happens when AI services sit at the end of the payment chain.
🔍 THE BOTTOM LINE
Zoli Rutter, a Sussex businessman, had been paying about £15 a month to use Claude for invoice analysis. On 19 June, Metro Bank texted him about a suspicious £90.90 charge. He replied “no” — it wasn’t his. The bank blocked that single transaction but did not freeze his account or card. Over the next 24 hours, fraudsters made dozens more charges, ranging from £90 to £200 each, buying Claude credits. Metro Bank eventually froze the card the following day. By then, £14,244 had gone through. The bank refunded Rutter only after The Guardian contacted them.
What Happened, Step by Step
Rutter had linked his Metro Bank debit card to his Anthropic account to pay the monthly Claude subscription. At some point, fraudsters gained access to his card details — Anthropic told The Guardian it found “no evidence” the compromise came from its own systems, suggesting the card details were stolen elsewhere.
The critical failure was Metro Bank’s response. When the bank’s fraud system flagged the first suspicious transaction and texted Rutter, he replied that it was unauthorised. The bank sent a follow-up text saying his account would be frozen. But it wasn’t — only the single flagged charge was blocked. The fraudsters then ran through dozens of subsequent transactions. Metro Bank says some were blocked by its systems, but many were not.
The bank eventually froze the card the next day. Rutter was left to fight for a refund. Metro Bank offered him £300 in compensation and told him, in a letter, that he had “not received the service we expect our customers to get.” Rutter says he plans to complain to the Financial Ombudsman Service.
The AI Angle: Claude Credits as Fraud Currency
This is not the first time Claude credits have been used in fraud. In May 2026, The Guardian reported on a US user whose financial details were used to buy Claude gift cards without authorisation. Reddit users have shared similar stories, suggesting a pattern of fraudsters targeting AI platform payment systems.
The reason is structural: AI credits are digital goods with instant delivery, no shipping address to verify, and easy resale. A fraudster who buys Claude credits with a stolen card can use them immediately or sell the account access. The transaction looks like a normal subscription payment to many bank fraud systems — which are tuned to detect card-not-present retail fraud, not recurring SaaS charges to an AI company that didn’t exist two years ago.
Anthropic says it has banned the user behind the fraudulent purchases and that anyone charged for a fraudulent purchase should contact its support team for a refund. Rutter has since been refunded by Anthropic, and Metro Bank’s temporary refund is being clawed back.
When the Bank Says “Frozen” but Means “One Charge Blocked”
The most troubling detail is the gap between what Metro Bank told Rutter and what it did. A text message saying his account “would be frozen” implied the flow of money would stop. Instead, only the single flagged transaction was blocked. The bank’s own fraud system then failed to catch dozens of subsequent charges to the same merchant — Anthropic — using the same card.
Metro Bank said in a statement: “This is a concerning case involving a third party that the customer had made previous genuine payments to and provided card details [to], which were then used to make a series of purchases without Mr Rutter’s knowledge.” The bank says it has since “taken steps to prevent delays in cards being blocked.”
The Financial Conduct Authority (FCA) — which, in a parallel development, recently endorsed Claude for use in its own AI innovation initiative — says unauthorised card fraud refunds should reach a customer’s account by the end of the next business day. Rutter’s refund took weeks and required press intervention.
The FCA’s Contradictory Position
The FCA’s dual role here is worth noting. On one hand, it is the regulator responsible for ensuring banks handle fraud cases like Rutter’s properly. On the other, it recently announced that Anthropic would provide Claude access to financial firms participating in an FCA AI initiative, as reported by The Guardian, to “help speed up their development work.” The same company whose payment system was used to defraud a bank customer is now being embedded in the UK’s financial regulatory infrastructure.
There is no suggestion of wrongdoing by either the FCA or Anthropic. But the proximity illustrates how quickly AI tools are moving from consumer product to regulatory instrument — and how the fraud risks to ordinary users have not yet been addressed with the same urgency.
NZ Angle: New Zealand Banks Face the Same Gap
New Zealand banks operate under similar card-fraud frameworks. The Banking Ombudsman Scheme handles disputes over unauthorised transactions, and NZ banks are subject to comparable refund obligations for card fraud. But the same structural gap applies: if a NZ customer’s card is used to buy AI credits by fraudsters, the bank’s fraud system may not flag recurring SaaS charges to a new AI company as suspicious — especially if the customer previously made legitimate payments to the same provider.
New Zealand’s Privacy Commissioner has previously flagged AI-related data risks, and the Privacy Act 2020 covers unauthorised use of personal financial data. But the enforcement mechanism for card fraud remains bank-driven, and as this case shows, bank fraud systems can fail even when the customer does everything right.
This case also connects to broader questions about AI and financial surveillance. We’ve previously covered how central banks are deploying AI in ways that create opaque economic systems, and how APRA is pushing Australian banks to manage AI risks. The Metro Bank case shows the flip side: banks’ own systems are not yet equipped to handle the fraud patterns that AI platforms create.
❓ FAQ
Can this happen with a New Zealand bank? Yes. NZ banks use similar fraud-detection systems. If your card is used to buy AI credits by someone else, the bank may not flag it as unusual — especially if you’ve previously paid for the same service. Contact your bank immediately if you see unauthorised charges.
What should I do if my card is used to buy AI credits fraudulently? Contact your bank immediately and ask for a refund under unauthorised transaction rules. Also contact the AI provider’s support team — Anthropic says it refunds fraudulent charges. Document everything, including any text messages from the bank.
Is Anthropic’s payment system secure? Anthropic says it has no evidence the card details were stolen from its systems. The fraud appears to involve card details stolen elsewhere and then used to buy Claude credits. But the pattern of Claude-specific fraud — multiple cases across months — raises questions about why AI credit purchases are attractive to fraudsters.
Did Metro Bank break the rules? The FCA says unauthorised card fraud refunds should be in a customer’s account by the end of the next business day. Rutter’s refund took weeks and required press involvement. Whether this constitutes a regulatory breach is a question for the Financial Ombudsman, where Rutter has filed a complaint.
🔍 THE BOTTOM LINE
The Metro Bank case is not really about AI — it’s about a bank that told a customer his account was frozen and then let £14,000 walk out the door. But AI is the reason the fraud worked: Claude credits are instant, digital, and resellable, making them an ideal target for card fraud. As AI platforms become payment endpoints, banks’ fraud systems need to catch up. Until they do, the gap between “we’ve frozen your account” and what actually happens will be measured in thousands of pounds.
