Customers utilizing artificial intelligence services from prominent providers like Anthropic and OpenAI have reportedly been overbilled by an estimated $1.7 million. This significant sum, identified through an audit by a specialized startup, highlights growing concerns over transparency and accuracy in AI service pricing.
Audit Uncovers Widespread Overcharges
Vaudit, a startup specializing in AI bill auditing, reviewed $34 million in invoices from 60 companies between March and June 2026. Their analysis revealed substantial overcharges, with the majority linked to Anthropic's Claude Code, though issues involving OpenAI's services were also found.
Among the companies reportedly utilizing Vaudit's services are major players like Panasonic, HP, and Honda, underscoring the broad impact of these billing discrepancies across various industries.
Causes of Billing Errors
The complexity of AI service billing, which often depends on factors such as model type, token usage, and API calls, makes independent verification challenging. However, one common cause of overbilling identified by Laura Bratton from The Information, was AI agents repeatedly attempting and failing a task, with customers being charged for each unsuccessful retry without realizing it.
Providers Issue Refunds Amidst Denials
Initially, both Anthropic and OpenAI reportedly denied evidence of billing errors. However, Vaudit later disclosed that after customer complaints, several AI providers, including Amazon, Google, Microsoft, Anthropic, and OpenAI, refunded approximately 80% of the disputed charges.
Despite these issues and the substantial costs, companies continue to invest heavily in premium AI models. The incident, however, emphasizes the critical need for more transparent and accurate billing practices as AI compute demands and prices continue to escalate across the industry.