[Bug] Anthropic API Error: False positive Usage Policy flag on legitimate base64-encoded CSV content

Bug Description
Bug Report: Legitimate business content falsely flagged as Usage Policy violation, session halted Date: 2026-07-01 Request ID: req_011CcajNParybk2Whd9zAZ88 Error message: API Error: Claude Code is unable to respond to this request, which appears to violate our Usage Policy (https://www.anthropic.com/legal/aup). Please double press esc to edit your last message or start a new session for Claude Code to assist with a different task. Steps to reproduce 1. Claude Code was editing a local CSV spreadsheet — a language-review questionnaire for a Guatemalan Spanish teacher, asking her to confirm regional naturalness/usage for a set of blog-article Spanish phrases (a content-quality workflow). 2. To create a new Google Sheet from this CSV, the CSV was encoded with base64 -i <file> -o <file>.b64. 3. The Read tool was used to open the resulting .b64 file (intending to pass its content as base64Content to the Google Drive MCP create_file tool). 4. The Read tool returned a truncation warning ("line too long to paginate, truncated at the 25,000-token cap") along with a partial view of the base64 string — this part was expected/normal behavior. 5. Immediately after, before the next tool call could proceed, the API Error above fired and the session became unresponsive. Likely cause of the false positive - The actual file content was entirely legitimate business data: a Spanish-language review questionnaire (102–104 items) asking a native-speaker teacher to confirm regional naturalness of phrases in published SEO blog articles. - One item in the list referenced a Guatemala travel-safety-advisory article's description of a land dispute in San Marcos department, questioning whether the article's claim of "cultivo de marihuana" (marijuana cultivation) should instead read "cultivo de amapola" (poppy cultivation) per reference sources — i.e., a factual accuracy correction request, not content promoting or instructing illegal activity. The item explicitly told the reviewer it was fine to decline answering if she felt uncomfortable. - Either that drug-related terminology in context, or the raw (non-human-readable) base64 blob itself, likely tripped an upstream safety classifier incorrectly. - The content does not constitute harm facilitation or policy violation in any reasonable reading — this appears to be a clear false positive. Impact - The task was forcibly halted right before completion (immediately before creating the new Google Sheet), requiring double-esc to edit the last message or starting a new session. - A routine, legitimate file-handling operation (base64-encoding a file and preparing to upload it) was blocked.

Environment Info

• Platform: darwin
• Terminal: Apple_Terminal
• Version: 2.1.197
• Feedback ID: 16f37b6e-4ddb-4d61-8125-fde411096c28

Errors

[]