Anthropic’s $16.6M Billing Error: Blocked Cards and 18 Emails Later
A South Korean Claude user received erroneous invoices totalling $16.6M, had his credit card blocked, and spent four days sending 18 emails to resolve it.
A South Korean user on Anthropic's free tier received two erroneous Claude billing invoices on 7 July, the first for $1.67 million and a second for $16.6 million (about 22.8 billion won) arriving less than 24 hours later. The charge attempts reached his bank and got his primary credit card blocked. Even after Anthropic admitted the mistake and confirmed no money was collected, the user spent four days and 18 emails trying to get written confirmation that the invalid invoices had been cancelled and his account cleared.
What happened
| Detail | Fact |
|---|---|
| First invoice (7 July) | $1.67 million (approx. £1.23M) |
| Second invoice (within 24 hours) | $16.6 million (approx. £12.2M / ₩22.8B) |
| User’s subscription level | Free tier, no billable API usage |
| Payment processor | Stripe |
| Emails sent to support | 18 over four days |
| Money actually collected | $0 |
| Root cause (per Anthropic) | Incorrect auto-reload setting |
The incident started on 7 July when the user received a failed payment notice for roughly $1.67 million. He was on Anthropic’s free tier and had no active API keys generating charges. Less than a day later, a second invoice arrived for $16.6 million. Both notices came through Anthropic’s official billing system and Stripe, so the user first checked his own AI automation projects, scripts, scheduled tasks, and account credentials. He found nothing that could explain the amounts.
As a precaution, he cancelled his Claude Max subscription and suspended the payment card linked to his account. Despite no money leaving his bank, the repeated charge attempts flagged by his card issuer were enough to get his primary credit card blocked.
How Anthropic responded (and how long it took)
After the user sent approximately 18 emails over four days, Anthropic escalated the case. Notably, the notification telling him a human would review the matter arrived from Anthropic’s Fin AI Agent, not a human support agent. The AI system informed him the issue had been forwarded to the company’s Privacy Team.
Anthropic eventually emailed the user directly. The company confirmed that an incorrect auto-reload setting had generated the invalid payment requests. It said the setting had been disabled, the billing configuration restored, and that no funds were collected. The email stated: “No money left your account. Our payment processor attempted a charge at the invalid amount and it was declined. Nothing was collected, and you owe nothing.”
Anthropic also confirmed the incident “was not the result of unauthorized access” and suggested the user re-enable auto-reload if he wanted automatic API credit top-ups to resume.
Why it matters
This is not a story about money being stolen. It is a story about what happens when automated billing goes wrong at scale and support processes fail to keep up. A free-tier user with zero API usage received invoices totalling more than $16 million. The invoices triggered real-world consequences: a blocked credit card and days of unresolved anxiety, even though the underlying error was entirely on Anthropic’s side.
The user’s frustration centred on the pace and quality of the response. “I still think it wasn’t handled properly,” he wrote on Threads. He also questioned why an automated email was sent after an error of this size: “I don’t understand how they could release an automatic email after causing a billions-won invoice error.”
For businesses and developers using paid AI API services, the episode is a reminder that billing automation can misfire in ways that affect real bank accounts and credit scores, even if the charge ultimately fails. Anyone running AI integrations tied to a business credit card should have alerts set for any charge attempt above a defined threshold.
Our take
Anthropic’s explanation is plausible. A misconfigured auto-reload setting generating runaway invoice amounts is an engineering bug, not fraud. What is harder to defend is the support experience. Eighteen emails over four days to confirm a company-acknowledged error is a poor result. Routing an escalation through an AI agent that then hands off to a Privacy Team adds friction exactly when a customer needs clarity fast.
The broader point is that as AI companies push API access to more users, billing infrastructure has to scale at the same rate. Incidents like this damage trust in a way that a bug fix alone does not repair. We have seen similar patterns in coverage of Anthropic’s communications around unverified model releases: the gap between what the company knows and what it tells users in real time tends to be wider than it should be.
If you are building products on top of any AI API, set hard spend limits at the API level and treat billing alert emails with the same urgency as a security alert. Do not rely on the provider’s support queue to be your safety net.
What to do about it
- Set a hard monthly spend cap inside your AI provider’s billing console, not just a soft alert.
- Use a virtual or single-use card for API subscriptions so a runaway charge attempt cannot block your main business card.
- Enable bank-side transaction alerts for any charge above a threshold you define.
- Document every support email with timestamps. If an error is company-acknowledged, reference that acknowledgement in every follow-up to move cases faster.
- Review your workflow automations regularly to confirm which ones hold live API credentials, so you can rule out your own tooling quickly if something like this happens.
A misconfigured auto-reload setting caused real disruption for a free-tier user. Keep your billing guardrails tighter than you think you need them.
Frequently asked questions
Did Anthropic actually charge anyone $16.6 million?
No. Anthropic confirmed that its payment processor Stripe attempted the charge but it was declined. No money left the user's account. The invoices were generated by an incorrect auto-reload setting and have since been cancelled.
What caused the Anthropic billing error?
According to Anthropic's email to the affected user, an incorrect auto-reload setting generated invalid payment requests. The company disabled the setting, restored the billing configuration, and confirmed no funds were collected.
Can a failed charge attempt block your credit card?
Yes. Even if no money is collected, repeated large charge attempts flagged by a payment processor can trigger fraud detection on your bank's side, resulting in a card block. This is what happened to the South Korean user in this incident.
How long did it take Anthropic to resolve the billing complaint?
The user reported sending approximately 18 emails over four days before Anthropic provided written confirmation that the invoices were cancelled and his account was cleared.