Anthropic billed a free-tier Claude user $16.6 million by mistake

Anthropic billed a free-tier user $16.6 million, by mistake.

The phantom invoice

A Claude user in South Korea got a failed-payment notice on July 7, 2026 for about $1.67 million. Less than a day later, a second invoice arrived: $16.6 million, according to the International Business Times UK. Anthropic later confirmed it was a billing error caused by an incorrect auto-reload setting, and that the user owes nothing.

No money actually left the account. The payment processor attempted the charge and it was declined, which is what you would expect when a consumer credit card meets an eight-figure invoice.

The real damage

The user still paid a price, just not in dollars. The repeated charge attempts reached the bank and got the user’s primary credit card blocked. Getting everything sorted took 4 days and about 18 emails.

There was one more twist in the support saga: even the message promising that a human would review the case came from Anthropic’s Fin AI support agent. An AI company, a seven-figure phantom invoice, and an AI apologizing for it. 2026 in one support ticket.

What Anthropic has and has not explained

Anthropic says the faulty auto-reload setting was disabled as a precaution. What the company has not explained is how the number climbed that high: an incorrect auto-reload might plausibly trigger an unwanted charge, but the leap from a free tier to $1.67 million, and then to $16.6 million within a day, remains unaccounted for publicly.

For anyone using AI services with saved payment methods, the practical lesson is unglamorous but real: check your auto-reload and spending-limit settings. Billing systems at AI companies are scaling as fast as everything else, and this case shows the failure modes can scale with them.

The user walked away owing nothing. The blocked card, the 4 days and the 18 emails were the cost of proving it.