Anthropic confirmed a massive billing error after its system tried to charge a South Korean free-tier user $16.6 million, despite zero API usage and no registered card.

The phantom invoices blocked his credit card and triggered a four-day support ordeal before the company responded.

Inside Anthropic’s $16.6 Million Phantom Invoice

A phantom invoice is a payment demand generated by a billing system without any underlying transaction or usage. That is exactly what a Korean developer, known as remy_notes, received from Anthropic in early July. The story quickly went viral after he shared the screenshots on Threads.

The first invoice arrived on July 7, totaling $1,669,875. Less than 24 hours later, a second one multiplied the figure almost tenfold to $16,627,739. The escalation occurred without any activity being recorded on the accoun

Follow us on X to get the latest news as it happens.

The developer initially suspected phishing. However, the emails came from Anthropic’s official domain and used the company’s legitimate Stripe infrastructure.

Furthermore, his dashboard showed zero API usage, no billable keys, and no payment method on file. He audited his own automation scripts and AI agents, finding nothing that explained the charges.

His bank declined two overseas charge attempts on July 8 because the amounts exceeded the card’s per-transaction limits. The repeated attempts still resulted in his primary credit card being blocked.

What Caused the Billing Error, According to Anthropic

Anthropic confirmed the mistake on July 12 and assured that no money was collected. The company attributed the invoices to an incorrect automatic credit reload setting, which it disabled as a precaution.

It also clarified that the incident did not involve unauthorized access. Anthropic has not explained how the reload value reached such an extreme figure.

The resolution, however, was far from smooth. The user spent four days and around 18 emails seeking written confirmation that the invoices were void. Meanwhile, automated replies kept arriving instead of human responses. He eventually asked the company to respond by Monday afternoon, Korean time.

The case also exposed a wider industry problem. Audit startup Vaudit reviewed $34 million in AI invoices from 60 enterprise clients and found roughly $1.7 million in overcharges. That represents an error rate of about 5%, much of it tied to AI billing systems. Clients affected by those errors included major corporations like Panasonic, HP, and Honda.

For users, the lesson is simple. Reviewing billing dashboards, monitoring card notifications, and reporting unusual charges quickly can prevent a glitch from becoming a financial nightmare.

Subscribe to our YouTube channel to watch leaders and journalists provide expert insights.

Источник: BeInCrypto