A recent audit conducted by the startup Vaudit has uncovered significant billing errors, revealing that 60 companies were mistakenly charged approximately $1.7 million for AI services, constituting around 5% of the total expenditure reviewed. The majority of the incorrect invoices were related to Claude Code. Notable clients of Vaudit include major corporations like Panasonic, HP, and Honda.
Auditors have identified three primary reasons for the extra charges. In certain instances, an outdated and less expensive model was used instead of the intended, more costly one. In other cases, charges were incurred for requests that were either incomplete or resulted in errors. Additionally, there are occasions where an agent repeatedly attempts a task that ultimately fails, leading to excessive token consumption and increased billing.
Anthropic has stated that it does not charge for incomplete or erroneous requests and believes that widespread overbilling is not a prevalent issue. OpenAI has also reported that they found no evidence of such problems among their client base. Despite the possibility that these discrepancies might originate from a third-party cloud provider, Vaudit has indicated that approximately 80% of the disputed charges have already been refunded to its clients.
Have you ever experienced incorrect charges for AI services? 🔥 — Yes, it's frustrating 🤔 — No, never @hiaimediaen
Informational material. 18+.
