Anthropic cancelled the wrong org, kept ~$3,900 of what we paid, and support says the decision is "final"

Detailed Analysis

A Reddit post from a business customer running a 30+ seat Claude subscription has surfaced as a cautionary account of Anthropic's billing and support operations breaking down under a compounding error. According to the poster, a duplicate organization was erroneously created and charged in parallel with their real, active org — a bug on Anthropic's side rather than a customer mistake. When the customer initiated a chargeback against the duplicate, Anthropic's system reacted by cancelling the wrong organization, terminating the legitimate subscription with roughly 315 of 365 paid days still unused. The customer repurchased immediately to avoid disrupting their team, then continued adding seats to the reinstated org as their company grew, accumulating roughly CA$9,900 in invoices. When that real org was also cancelled, Anthropic refunded only the base subscription invoice (~CA$5,000) and kept the remaining ~CA$4,900 tied to the added seats, despite the customer providing a detailed spreadsheet breakdown and repeatedly offering to accept Anthropic's own proration math if applied correctly.

The core of the dispute is a billing logic error: Anthropic appears to be crediting one refund against two separate charged entities — the duplicate org and the real org — effectively double-counting a single refund to claim the customer has been "made whole." The customer's calculation of unjust enrichment is straightforward: money was paid, service was not received in return, and repeated attempts to correct the record through support were met with form responses declaring the decision "final," an inability to obtain a case number after three requests, and confirmation that the original duplicate org — flagged for cancellation back in March — still hadn't been closed four months later. Notably, the customer says many support replies came from an AI agent rather than a human, and each exchange took weeks, suggesting an overwhelmed or under-resourced support pipeline rather than deliberate bad faith.

This matters because it exposes a gap between Anthropic's positioning as an enterprise-ready AI vendor and the operational maturity of its billing and account-management infrastructure. As Claude has scaled into business and enterprise deployments — competing directly with OpenAI, Google, and Microsoft for corporate AI budgets — the expectation from paying customers is that account administration, invoicing, and dispute resolution match the reliability of the underlying model. A duplicate-org billing bug, followed by an erroneous cancellation of the wrong account, followed by an unresolved seat-refund shortfall, represents exactly the kind of back-office failure that erodes trust with exactly the customer segment (multi-seat, annual-contract businesses) that generates the most durable revenue for AI labs. The customer's own admission — that they're wary of doing a second chargeback for fear of triggering account suspension — highlights a broader power asymmetry: enterprise customers who depend on a tool operationally have limited leverage to contest billing errors without risking service disruption.

The episode also fits a recurring pattern across the generative-AI industry, where rapid scaling of paid seats, tiered plans, and API/subscription hybrids has outpaced the growth of support and billing systems. Anthropic, OpenAI, and others have all faced customer complaints about opaque support processes, chatbot-mediated ticket handling, and slow escalation paths for account-level (rather than product-level) issues. As AI companies push further into enterprise sales — where multi-seat annual contracts, SOC2 compliance, and dedicated account management are typically table stakes — stories like this one signal that the back-office and customer-success layer of these companies may still be immature relative to their technical capabilities. For prospective business buyers evaluating Claude Team or Enterprise plans, the post serves as a specific, documented warning: technical excellence in the product does not guarantee commensurate reliability in billing, refunds, or dispute resolution once something goes wrong.