Anthropic's $16.6 Million Phantom Invoice: How a Free Claude User Got a Billing Shock

A report from ZDNet Korea, cited by the International Cyber Digest, provides a precise timeline: the first email arrived on July 7 at 10:20 PM — a "payment unsuccessful" notice for $1,669,875.90. A day later, on July 8 at 11:42 PM, a second notice arrived for $16,627,739.70 . The exact mechanism has not been publicly explained by Anthropic, but the user's account dashboard showed no corresponding usage data to justify any charge, let alone one of that magnitude .

Why the User Initially Suspected Phishing

When the invoice arrived via email from what appeared to be an Anthropic billing address, the user initially believed it was a sophisticated phishing attempt . The astronomical amount — coupled with the fact that they were on a free plan — made the invoice seem nonsensical as a legitimate charge. The user stated that they had never provided a payment method to Anthropic, making the invoice appear to be an obvious scam . Only after cross-checking with their actual Claude account dashboard and seeing the phantom invoice reflected there did they realize it was a genuine billing system error rather than an external scam.

How Card Charge Attempts Reached an Unregistered Card

The user reported that they had never registered a credit card with Anthropic . Despite that, Anthropic's billing system — likely through Stripe — attempted to process charges against a card that was not on file . The user shared evidence that the system was trying to collect payment even though no valid payment instrument existed in the account . How Stripe attempted to run charges without a stored card method has not been detailed, but it suggests either a billing system bug that generated charge attempts against an empty or corrupted payment field, or that Anthropic's billing pipeline fired collection attempts without properly checking whether a payment method was present .

What Anthropic Has and Has Not Said

Anthropic has provided only a standard privacy response — that it cannot discuss individual account details publicly . The company has not acknowledged any specific billing system bug, has not explained how a free-plan user generated a $16.6 million invoice, has not clarified whether Stripe configuration errors were involved, has not stated whether refunds or credits were issued, has not issued any public statement or postmortem about the incident, and has not addressed how charge attempts were made against an unregistered card .

Broader Billing Issues Raised Against Anthropic

The $16.6 million incident is the most extreme example in a pattern of billing problems that have plagued Anthropic throughout 2026.

$1.7 Million in Enterprise Overcharges

The auditing startup Vaudit, founded by former Oracle director Michael Hahn, examined $34 million in AI invoices from 60 companies (mostly for Claude Code) and identified roughly $1.7 million in overcharges . The audited client list included major corporations such as Panasonic, HP, and Honda . Vaudit found that Anthropic refunded some of the overbilled amounts but did not admit fault .

'Ghost Charges' on Claude Max Plans

Multiple Claude Max subscribers reported unexplained "Extra Usage" charges totaling $180 or more — with 16 back-to-back invoices in some cases — despite being nowhere near their computers. In one case documented on Wikidocs, a user was on a yacht in San Diego while $180 in phantom charges accumulated, and Anthropic support allegedly went unresponsive for over a month .

The 'HERMES.md' Billing Bug

In April 2026, a developer discovered that the text string "HERMES.md" appearing anywhere in local git commit history caused Claude Code to silently route API requests from a flat-rate Max 20x plan onto metered "extra usage" billing, draining $200.98 in a single day . Anthropic initially refused a refund, and the user had to reverse-engineer the trigger themselves . After public pressure, Anthropic acknowledged the error and confirmed it would issue full refunds and additional credits to affected users .

Unauthorized Gift Subscription Charges

In April–May 2026, multiple users reported being charged $200 for "Claude Max gift subscriptions" sent to randomly generated 27-character iCloud email addresses that bounced . The Guardian covered this as an emerging form of AI chatbot billing fraud, noting that affected users had their accounts suspended when they contacted Anthropic .

$500 Million Enterprise Billing Incident

Axios reported that an enterprise client accidentally generated a $500 million bill in a single month by failing to set usage limits on employees using Claude . An AI consultant told Axios that the enterprise gave its employees unrestricted access to Claude, and because enterprise AI services are typically billed based on token consumption rather than flat subscriptions, costs escalated rapidly .

Class-Action Lawsuit Over Rate Limits

In June 2026, Washington, D.C. customer Karl Kahn filed a federal lawsuit against Anthropic, alleging the company exaggerated the usage limits of its $200/month Max plans . The lawsuit, which seeks class-action status, claims Anthropic deceived customers regarding the usage restrictions of its high-end Max and Max 20x subscription tiers .

Persistent Support Failures

Users across Hacker News, Reddit, and community forums have reported waiting 30+ days for Anthropic support to respond to billing disputes . One user reported waiting over a month for a response to a billing issue , while another described getting only automated AI bot replies with no human follow-up .