Anthropic Acknowledges $16.6 Million Billing Error, User Reveals Repeated Charge Attempts Blocked Bank Card, Four-Day Support Ordeal

A South Korean Claude user says Anthropic's erroneous £12.2 million ($16.6 million, about ₩22.8 billion) billing demand led to repeated charge attempts that blocked his primary credit card, even after the company admitted the mistake.

He says he then spent four days trying to get a clear response from support as he sought confirmation that the invalid invoices had been cancelled and his account cleared.

The incident first came to light after the user, who said he was on Anthropic's free tier with no billable API usage, revealed invoices that escalated from about £1.23 million ($1.67 million) to £12.2 million ($16.6 million) within 24 hours.

Anthropic has since confirmed the billing error and said no money was taken, while the user's latest posts describe the disruption that followed and his efforts to resolve it.

'It Wasn't Handled Properly'

The user said the billing error became a customer support issue long after the invoices were issued.

In updates shared on Threads, he said he contacted Anthropic repeatedly over four days, sending about 18 emails as he sought confirmation that the charges had been cancelled and his account had been cleared.

He said repeated payment attempts reached his bank and resulted in his primary credit card being blocked, despite no money ultimately leaving the account.

'I still think it wasn't handled properly,' he wrote.

The user said he expected a more urgent response given the size of the erroneous invoices.

'It seems like the urgent fire has been put out, but personally, I don't understand how they could release an automatic email after causing a billions-won invoice error,' he wrote.

Even after the case was escalated, the notification advising that a human would review the matter came from Anthropic's Fin AI Agent, informing him that the issue had been forwarded to the company's Privacy Team.

Anthropic Admits Billing Mistake And Explains What Went Wrong

Anthropic later told the customer that an incorrect auto-reload setting had generated invalid payment requests.

In an email shared by the user, the company said it disabled the setting as a precaution, restored the account's billing configuration and confirmed that no funds had been collected.

'No money left your account,' Anthropic said. 'Our payment processor attempted a charge at the invalid amount and it was declined... Nothing was collected, and you owe nothing.'

The company also said the incident 'was not the result of unauthorized access' and advised the user to re-enable the auto-reload feature if he wished to continue using automatic API credit top-ups.

Invoice Jumped More Than £11 Million In Just 24 Hours

The dispute began on 7 July when the user received a failed payment notice seeking about £1.23 million ($1.67 million).

Less than a day later, a second invoice arrived requesting £12.2 million ($16.6 million, about ₩22.8 billion).

Because both payment notices came through Anthropic's official billing system and Stripe, the company's payment processor, the user said he initially investigated whether one of his own AI automation projects could somehow have generated the charges.

After inspecting his AI agents, automation scripts, scheduled tasks and account credentials, he said he found no billable Anthropic API keys or other evidence that could explain the invoices. As a precaution, he cancelled his Claude Max subscription and suspended the payment card linked to the account.

'I'll Follow Up'

The user said he plans to continue documenting the incident while waiting for additional responses from Anthropic.

In a Threads update posted after receiving the company's explanation, he said he had sent a total of 18 emails and asked Anthropic to respond by Monday afternoon Korean time.

'I have scheduled a meeting with the reporter, and I think I can tell you more details around Tuesday afternoon or Wednesday,' he wrote.

He also encouraged others to reference his experience, adding: 'If you don't mind, you can share the conversation or use the article as a reference... Just be careful not to distort the information.'