Anthropic Customers Find Errant Charges, Auditing Startup Says

2026-06-25 · Source: The Information · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cloud Computing & IT Infrastructure · Depth: Fundamental Awareness, short

Summary

Vaudit, an AI bill-auditing startup, identified approximately \$1.7 million in mistaken overcharges across \$34 million in bills from 60 companies for Anthropic and OpenAI products between March and June. These inaccuracies, primarily involving Anthropic's Claude Code, affected customers like Panasonic, HP, and Honda. Common issues included billing for newer, more expensive AI models when older ones were used, charges for incomplete requests or error messages, and "retry storms" where failed AI agents continued processing. While Anthropic and OpenAI dispute the widespread nature of these problems, Vaudit reported that about 80% of the identified overcharges were credited back by major cloud providers and the AI firms. Vaudit, a 30-person startup founded in 2023, offers software to compare AI usage data with invoices, charging 1% of reviewed bills and 30% of refunds.

Key takeaway

For Directors of AI/ML or IT Professionals managing enterprise AI deployments, you should implement robust internal auditing processes for AI service bills. Given observed billing inaccuracies, proactively comparing your actual AI model usage and request completion rates against vendor invoices can prevent significant overcharges. Consider deploying third-party tools like Vaudit to automate this verification, ensuring cost efficiency and accurate budget allocation for your AI initiatives.

Key insights

AI enterprise billing is complex and prone to errors, leading to significant overcharges for customers.

Principles

• Enterprise AI billing is often opaque.
• Overcharges can occur from model misidentification.
• Failed AI requests may still incur costs.

Method

Vaudit's software, installed via SDK, captures raw AI usage data to compare against invoices and billing records for discrepancies.

In practice

• Audit AI bills for model usage discrepancies.
• Verify charges for incomplete AI requests.
• Monitor for "retry storms" in AI agents.

Topics

• AI Billing
• Cloud Cost Management
• Anthropic Claude
• OpenAI
• Vaudit
• Enterprise AI Auditing

Best for: CTO, VP of Engineering/Data, Entrepreneur, Director of AI/ML, Consultant, IT Professional

Related on AIssential

• Anthropic Users Are PISSED — Anthropic's billing system penalized users for mentioning specific third-party tools, leading to unexpected charges and service denials.
• ChinAI #354: Industry Gossip - overdue training fee payments and overhyped embodied AI — China's AI sector faces significant financial and operational hurdles, including debt, overhyped capabilities, and failed acquisitions.
• Microsoft Cancels Internal Anthropic Licenses As Shift To Token-Based AI Billing Blows Up Annual Budgets In Months — Token-based AI billing models are proving prohibitively expensive for large-scale enterprise adoption, leading to rapid budget depletion.
• Anthropic backs off unpopular billing overhaul as price war with OpenAI looms — Market competition, regulatory scrutiny, and investor sentiment directly influence AI platform billing strategies.
• Anthropic "pauses" token-based billing for its Claude Agent SDK — AI platform providers may adjust billing models in response to user impact and strategic goals.
• Every Hourly Billing Flaw Explained in Detail (And Why It Can’t Survive AI) — AI's efficiency gains are making the traditional hourly billing model unsustainable for professional services.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Information.