Anthropic Claude Code HERMES.md billing flaw (consumerrights.wiki)

Anthropic has announced a significant billing flaw within its Claude Code product, which inaccurately charged users direct API fees instead of adhering to their flat-rate subscription plans. This technical error was triggered specifically by the string "HERMES.md" found in users' Git commit histories, leading one subscriber to incur over $200 in charges despite having substantial capacity remaining on their $200 monthly Max plan. The flaw arises during Anthropic's recent enforcement against third-party agent harnesses like the open-source Hermes Agent from Nous Research, which now faces additional billing complexities as subscribers must navigate a pay-as-you-go model for using these tools. This incident highlights critical implications for the AI/ML community as it underscores the challenges of integrating third-party tools with existing subscription models. The automated detection mechanism appears to mistakenly associate certain string inputs with unauthorized use, leading to unintentional billing consequences. Furthermore, Anthropic's response to user support inquiries, opting not to issue refunds for the error and branding it as non-refundable, raises concerns about user trust and transparency in billing practices within AI services. This situation serves as a cautionary tale about the need for clearer communication and robust error-checking in automated systems, particularly as the integration of AI tools becomes increasingly prevalent in software development workflows.