Anthropic mistakenly billed a free user $16.6 million due to a system error, leading to confusion and a subsequent confirmation from the company.
In a surprising turn of events, AI startup Anthropic accidentally billed a user $16.6 million for services they had not used. The developer, known by the pseudonym remy_notes, reported the issue on Threads, revealing that he initially received an invoice for $1.6 million on July 7, which he suspected was a phishing scam. However, shortly after, he received a second invoice demanding an even larger payment.
Despite the invoices appearing legitimate, with links to Anthropic’s official domain and payment processor Stripe, remy_notes' account showed no API usage or billable keys. On July 8, his bank card was declined for two payment attempts, each exceeding his transaction limit, with the charges listed under the name ANTHROP.
The developer investigated his system thoroughly but found no explanation for the erroneous charges. To be safe, he canceled his Claude Max subscription and blocked his payment card. After extensive communication with customer support, which included automated responses from an AI named Fin, the situation was escalated to Anthropic's privacy department.
On July 12, Anthropic confirmed the billing error, attributing it to a misconfigured auto-replenishment feature. They reassured remy_notes that no funds had been withdrawn and that the feature had been disabled as a precaution. However, they were unable to clarify why the system generated the specific amount of $16.6 million.
Broader Issues with Billing
This incident is not an isolated one; reports from the auditing firm Vaudit indicated that corporate clients of Anthropic and OpenAI faced similar overcharges, totaling $34 million across 60 organizations from March to June. Approximately 5% of these invoices were found to contain excessive charges due to billing errors, including incorrect rates for older models and charges for failed requests.
After inquiries from Vaudit and affected clients, refunds began to flow, with around 80% of the overcharges being reimbursed by major companies like Amazon, Google, Microsoft, Anthropic, and OpenAI. Earlier in the spring, Anthropic had also faced backlash over limitations on its Claude Code service, prompting an increase in those limits after negotiations with SpaceX.
