In early July 2025, an incident occurred that forced many to reconsider their attitude toward automated payment systems in the field of artificial intelligence. A South Korean developer under the pseudonym remy_notes, using Anthropic's free plan, received a bill from the company for an astronomical amount of $16.6 million. This happened despite the fact that he did not perform any operations through the API, did not link a bank card, and did not generate any tokens.
Timeline of the "Phantom" Bill
The first bill for $1.67 million arrived on July 7. A day later, the amount increased tenfold, reaching $16,627,739. The developer checked his scripts and AI agents but found no reason for the charge. The bank rejected the debit attempts twice because the amounts exceeded the card limits, but after several failed transactions, the card was blocked. Resolving the issue took four days and required about 18 emails to customer support, which only responded with automated replies.
Anthropic's Official Position and the Broader Problem
The company acknowledged the error, explaining it as a glitch in the auto-recharge credit settings. The feature has since been disabled. However, the key question remains unanswered: why did the auto-recharge amount reach such absurd values? No data leaks or third-party access to the service were detected.
This case is just the tip of the iceberg. The startup Vaudit conducted an audit of $34 million in AI service bills for 60 major clients, including Panasonic, HP, and Honda. The results are shocking: overpayments amounted to about $1.7 million (approximately 5% of the total amount). The vast majority of errors are related to billing for AI-based solutions.
Conclusions and Recommendations
This incident reveals a systemic vulnerability in the payment mechanisms of major AI providers. Users should take note: regularly check payment panels, monitor bank notifications, and respond promptly to suspicious charges. An accidental glitch can turn into a serious financial problem if measures are not taken in time.
Expert opinion: The story with Anthropic is not an isolated bug but a symptom of the growing complexity of billing systems in the era of widespread AI adoption. While companies focus on developing models, the security and transparency of payments remain a secondary concern. This could become a "time bomb" for the entire industry if strict standards for auditing and user control are not introduced.