An incident involving a $16.6 million invoice sent to a user of the free Claude tier has exposed serious gaps in the billing systems of AI giants. A developer from South Korea, known under the pseudonym remy_notes, encountered this in early July, and this story is not an isolated case but a symptom of a larger problem.
It all started on July 7, when remy_notes received the first invoice for $1.6 million. Initially, he suspected phishing, but the next day a second letter arrived — this time for $16.6 million. All links led to the official Anthropic domain and the Stripe payment system. Meanwhile, the dashboard showed zero API usage, and there were no billable keys on the account. The developer's bank card twice declined the debit attempts, as the amounts exceeded the per-transaction limit. The transfers came from the U.S. merchant ANTHROP.
The user thoroughly checked all processes on his Mac mini, including eight AI agents, but found nothing that could explain the incident. As a preventive measure, he canceled his Claude Max subscription and blocked his card. However, it took four days and about 18 emails to support to clarify the circumstances. Some responses came automatically — from an Anthropic AI agent named Fin. Only on July 12 did the company confirm the error, stating that not a single dollar had been debited. The cause was attributed to a misconfigured auto-reload balance feature, which was disabled as a precaution. Anthropic also emphasized that the account had not been hacked, but support could not explain why the system generated exactly $16.6 million.
A Systemic Problem, Not an Isolated Glitch
This case is just the tip of the iceberg. Audit firm Vaudit, reviewing invoices totaling $34 million across 60 organizations from March to June, identified overcharges amounting to approximately $1.7 million (roughly 5% of the audited volume). Most invoices were related to the use of Claude Code by Anthropic. Specialists discovered several systemic errors: clients were billed at rates for new models, even though the cheaper older version was actually used; charges were recorded for requests that ended in errors or produced no results. A separate category consisted of so-called "retry storms" — a situation where an AI agent repeatedly retries a failed task in the background, with each attempt reflected on the invoice.
After Vaudit and clients raised concerns, companies began issuing refunds. About 80% of the overpayments were compensated by Amazon, Google, Microsoft, Anthropic, and OpenAI.
My analysis: This incident highlights a critical vulnerability in the infrastructure of AI services. When AI agents manage billing, and code errors can lead to multi-million dollar invoices, user trust is undermined. Companies need to implement stricter transaction verification mechanisms and human oversight, especially when processing large sums. Otherwise, such "glitches" will become the norm rather than the exception.