Anthropic's Claude Code agent triggers billing errors on specific keywords

Anthropic's Claude Code agent is exhibiting a bug where specific substrings in commit messages or file content trigger a sensitive routing mechanism, leading to unexpected charges or refusals. This issue, first reported in April, has surfaced in multiple instances, including keywords like "HERMES.md" and "openclaw.inbound_meta.v1", and a system reminder about malware analysis. The problem causes the agent to incorrectly route requests to a more expensive tier or refuse to process them, even when the content is benign. AI

IMPACT This bug highlights potential issues with LLM-based tools, impacting user trust and costs, and underscoring the need for robust error handling and transparent billing.

RANK_REASON The cluster describes a bug in a specific product (Claude Code) that affects its functionality and billing, rather than a new model release or fundamental research.

Read on dev.to — LLM tag →

• Anthropic
• bcherny
• Boris Cherny
• Claude Code
• claude-opus-4-6[1m]
• GitHub
• Hacker News
• HERMES.md
• openclaw.inbound_meta.v1
• sasha-id
• t3.gg
• Theo Browne
• X

• product
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →