Anthropic, the developer of the well-known Claude model, has found itself at the center of an awkward incident. Due to a glitch in the payment system, a user on the free tier from South Korea received a bill for an astronomical amount — over $16.6 million. The situation was compounded by the fact that the affected user had not linked a bank card or made any transactions via the API.
Timeline of the "Phantom" Bill
In early July, a developer under the pseudonym remy_notes published screenshots of emails from Anthropic on Threads. The first notification arrived on July 7 for $1,669,875. Less than a day later, the amount increased tenfold — to $16,627,739. Meanwhile, in the user's personal account, API usage remained at zero, no paid keys were present, and no card was linked.
At first, remy_notes thought it was phishing. However, the emails came from Anthropic's official domain via the Stripe payment service. The bank rejected two attempts to debit the funds on July 8, as the amounts exceeded the card's limit. But after several failed attempts, the primary card was eventually blocked.
Official Response from Anthropic
The company acknowledged the error and stated that the money was not actually debited. The cause of the glitch was attributed to a malfunction in the auto-recharge credit settings. This feature has since been disabled "for safety reasons." However, Anthropic did not explain why the auto-recharge amount reached such absurd values.
Resolving the situation took four days. To obtain written confirmation of the invoice cancellation, the developer had to send about 18 emails, which only received automated replies.
The Problem Is Broader Than It Seems
This is not an isolated case. The startup Vaudit reviewed $34 million in invoices for AI services from 60 large clients and discovered overpayments of approximately $1.7 million — about 5% of the total amount. The largest errors were attributed to billing for AI-based solutions. Among the affected companies are Panasonic, HP, and Honda.
My comment: This incident highlights a critical vulnerability in the billing infrastructure of AI companies. While clients risk encountering "phantom" bills and support teams spend days resolving the issue, trust in the service is undermined. I recommend that all users of AI platforms regularly check their payment panels and set strict limits on auto-recharge. A single bug in the configuration could cost you not only your peace of mind but also the blocking of your bank card.