HERMES.md: Anthropic bug causes $200 extra charge, refuses refund
Anthropic's Claude Code is facing heat after a bizarre bug caused users to be billed extra for having the string "HERMES.md" in their git commit history. The company's alleged refusal to refund for "technical errors" has ignited outrage among users, leading many to advise chargebacks and consider abandoning the platform. This incident highlights the challenges of billing transparency and customer support in the rapidly evolving AI services landscape.
The Lowdown
A peculiar and costly bug in Anthropic's Claude Code has been uncovered, leading to unexpected charges for users. The core issue revolves around a highly specific string within git commit messages, which silently redirects API requests to an 'extra usage' billing tier instead of a user's existing plan quota.
- The Bug: If the case-sensitive string 'HERMES.md' is present in a recent git commit message, Claude Code routes API requests to extra usage billing. This occurs regardless of whether a file with that name exists on disk.
- Mechanism: Claude Code incorporates recent commit history into its system prompt. A server-side mechanism seemingly misinterprets the presence of 'HERMES.md' in this context, triggering an alternative billing pathway.
- Impact: The discoverer of the bug incurred over $200 in charges while their Max plan quota remained largely unused. The bug also rendered projects unusable upon depletion of these extra credits, with misleading "out of extra usage" error messages hindering diagnosis.
- Trigger Specificity: The bug is case-sensitive and requires the ".md" extension. Variations like 'hermes.md', 'HERMES', or 'HERMES.txt' do not trigger the issue.
This incident underscores the hidden complexities and potential financial pitfalls in integrating AI tools with development workflows, especially when billing logic is tied to unexpected content patterns.
The Gossip
Refund Refusal Ruckus
The most prevalent theme revolves around Anthropic's controversial stance on not compensating for 'technical errors' or 'degraded service.' Many commenters express shock and disbelief at this policy, deeming it unacceptable for a legitimate business. Several users share their own negative experiences with Anthropic's customer service, including double billing and difficulty obtaining resolutions, leading some to resort to credit card disputes.
Legal Loopholes & Chargeback Calls
A significant portion of the discussion centers on the legal and ethical implications of Anthropic's billing practices and refusal to refund. Commenters question the legality of such a policy in various jurisdictions and suggest that it opens the door for companies to intentionally overcharge. Many advocate for users to initiate credit card chargebacks as a practical and effective means to recover incorrectly billed funds, noting that such disputes are often successful when the company acknowledges the error.
AI-Assisted Apologies (or Lack Thereof)
There's widespread speculation that Anthropic's unhelpful and tone-deaf customer support responses, particularly the one refusing a refund, were themselves generated by an AI (ironically, perhaps by Claude). Commenters find this highly ironic given the context of the bug, perceiving it as a preview of a frustrating future where automated support systems are designed to deny accountability. This fuels further cynicism about Anthropic's customer-centricity.
Customer Exodus & Open-Source Options
Many users express deep dissatisfaction with Anthropic's recent performance and customer handling, indicating a significant erosion of goodwill. Several commenters state their intention to cancel subscriptions or have already switched to competing models and services. There's a strong sentiment that such issues make proprietary LLMs less appealing, driving users towards local and open-source alternatives as a more reliable and controllable option.