Anthropic tried to bill a free-tier user $16.6M

A South Korean developer on Anthropic's free plan — with zero API usage and a dashboard reading $0.00 — got hit with a $16.6 million invoice. A day earlier the same bill was $1.67M, meaning it ballooned roughly 10x overnight. He assumed phishing until he traced the sender and payment link back to Anthropic's official domain. His bank killed the charge for blowing past the card's per-transaction limit.

It's not a one-off. Auditing startup Vaudit says it found ~$1.7M in overcharges across $34M of AI invoices from 60 companies (March–June), most tied to Claude Code — flagging wrong-model pricing, charges for failed requests, and background retry storms quietly running up bills. Anthropic disputes the findings, but ~80% of contested charges were later credited back.

The glitch lands amid months of GitHub reports about contradictory Anthropic billing emails.

Why it matters: as AI usage-based billing scales, a broken invoice pipeline can turn a $0 account into a life-altering charge.