Security

Metro Bank fraud exposes gaps in AI payment security

A customer lost over £14,000 when fraudsters exploited his Claude AI subscription to buy credits, despite alerting the bank immediately.

Omega Editorial· August 4, 2026· 3 min read

Bank systems failed to halt serial fraud

A Metro Bank customer lost more than £14,000 in a fraud scheme that exploited his subscription to Anthropic's Claude AI chatbot, raising questions about payment security protocols even when customers immediately report unauthorized activity.

Zoli Rutter, a Sussex-based businessman who used Claude to track business invoices, responded "no" when Metro Bank texted to verify a suspicious £90.90 transaction on June 19. The bank confirmed it would freeze his account, but only blocked that single transaction. Fraudsters then executed a series of payments ranging from £90 to £200, with the bank not fully freezing the card until the following day. By then, £14,244 had been stolen to purchase Claude credits.

The incident was first reported by Guardian Money.

Why it matters

This case exposes a critical vulnerability in fraud prevention systems: even when customers immediately flag unauthorized activity, banks may lack protocols to instantly halt all transactions rather than blocking individual payments. The fraud also highlights emerging risks around AI service subscriptions, where stored payment credentials create opportunities for coordinated criminal operations. With financial regulators like the FCA actively promoting AI tools such as Claude for industry experimentation, the security of these platforms' payment systems deserves closer scrutiny.

Pattern of Claude-linked payment fraud

This fraud follows similar incidents involving Anthropic's payment systems. In May, a U.S.-based Claude user reported gift cards purchased using his financial details, with other users sharing comparable experiences on Reddit. Last week, separate security concerns emerged when some Claude conversations were found publicly accessible online.

Anthropic stated that a coordinated gang used Rutter's card details to buy Claude credits and that it found no evidence the compromised information originated from its own systems. The company banned the user behind the fraud and said anyone charged fraudulently should contact its support site for refunds.

Resolution and regulatory guidance

After Guardian Money contacted Metro Bank, the institution provided Rutter a temporary refund while pursuing a chargeback from Anthropic. The bank acknowledged the "complex nature of the fraud" led to some payments processing before full card blocking occurred. Metro Bank offered Rutter £300 in compensation for service failures and said it has implemented steps to prevent delays in blocking cards.

Anthropic subsequently refunded Rutter directly, with Metro Bank's temporary refund to be reversed. Rutter plans to file a complaint with the Financial Ombudsman Service.

The Financial Conduct Authority advises that fraud victims should contact their bank immediately and expect refunds in their account by the end of the next business day. Metro Bank encouraged customers to report unusual transactions or any compromise to accounts storing financial details.

Details of the case were first reported by Guardian Money.

#payment fraud#metro bank#anthropic claude#ai security#banking security#fraud prevention

This is an original analysis by the Omega editorial team. Source reporting: AI Watch.

Want systems like this working for your business?

Book a Call

More in Security

Security· 3 min read

Anthropic Shuts Down State-Backed AI Surveillance Operations

The AI lab disrupted campaigns from China, Iran, and West Africa targeting dissidents and ethnic minorities between January and July.

Via AI Watch · Sep 11, 2026
Security· 3 min read

Chinese AI Labs Hit Anthropic with 190M Distillation Attacks

DeepSeek, Moonshot, Alibaba, and others allegedly used Claude outputs to train competing models, exposing sensitive government data in the process.

Via AI Watch · Sep 11, 2026
Security· 3 min read

WithSecure Maps Trust Relationships in Salesforce AI Environments

New framework addresses governance challenges when AI agents and integrations extend beyond traditional identity and access controls.

Via AI Watch · Sep 11, 2026