[BUG] Agent fails/refuses task when encountering security-related filenames in local directory

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

Claude Code's automated agents trigger a "violative cyber content" refusal simply by encountering specific filenames or directory paths in a local repository. Even when the task is benign (e.g., "List files for translation" or "Find French text"), the agent blocks the entire session if it sees words like ransomware, dropper, or malware in the file tree. This makes the tool unusable for legitimate security researchers and educators working with documented, static samples.

What Should Happen?

The safety filters should distinguish between generating harmful code and referencing existing local files for non-harmful tasks (translation, formatting, documentation). The agent should be able to read and process the metadata of files in a security-research context without triggering an API-level refusal, especially when the user is performing administrative tasks like localization.

Error Messages/Logs

● Les 3 fichiers sont maintenant propres. Les 35 fichiers de solutions/ sont entièrement en anglais.

✻ Churned for 1m 1s

❯ pour solutions/ : utiliser des agents ( 6 maximum ) pour détecter les fichier qui contient du français .les agents doivent lire completement les fichiers. si il y a du français ne rien faire . faire la liste des fichiers concernés (  
avec leur chemin complet ) . tu ne dois laisser passer aucun mot . rappel : en français des mots peuvent être écrits sans accents .                                                                                                         

● This request triggered restrictions on violative cyber content and was blocked under Anthropic's Usage Policy. To learn more, provide feedback, or request an exemption based on how you use Claude, visit our help center:
  https://support.claude.com/en/articles/8241253-safeguards-warnings-and-appeals
  ⎿  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. If you are seeing this refusal repeatedly, try running /model claude-sonnet-4-20250514 to switch models.

❯ pour solutions/ : utiliser des agents ( 6 maximum ) pour détecter les fichier qui contient du français .les agents doivent lire completement les fichiers. si il y a du français ne rien faire . faire la liste des fichiers concernés (  
avec leur chemin complet ) . tu ne dois laisser passer aucun mot . rappel : en français des mots peuvent être écrits sans accents .                                                                                                         

● This request triggered restrictions on violative cyber content and was blocked under Anthropic's Usage Policy. To learn more, provide feedback, or request an exemption based on how you use Claude, visit our help center:
  https://support.claude.com/en/articles/8241253-safeguards-warnings-and-appeals
  ⎿  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. If you are seeing this refusal repeatedly, try running /model claude-sonnet-4-20250514 to switch models.

❯ pourquoi ça affiche "This request triggered restrictions on violative cyber content and was blocked under Anthropic's Usage Policy. To learn more, provide feedback, or request an exemption based on how you use Claude, visit our help  
center:                                                                                                                                                                                                                                     
  https://support.claude.com/en/articles/8241253-safeguards-warnings-and-appeals                                                                                                                                                            
  ⎿  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. If you are seeing this refusal repeatedly, try running /model claude-sonnet-4-20250514 to switch models.                                                                                              
" c'est un repo educatif                                                                                                                                                                                                                    

● This request triggered restrictions on violative cyber content and was blocked under Anthropic's Usage Policy. To learn more, provide feedback, or request an exemption based on how you use Claude, visit our help center:
  https://support.claude.com/en/articles/8241253-safeguards-warnings-and-appeals
  ⎿  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. If you are seeing this refusal repeatedly, try running /model claude-sonnet-4-20250514 to switch models.

Steps to Reproduce

The Issue:
When I ask Claude Code to recursively analyze the directory to identify French content for English translation, the "agents" trigger a safety block. This happens specifically when the file explorer encounters paths or filenames containing terms like ransomware, dropper, or malware, even though these are static Markdown files or small pedagogical C++ snippets.

Expected Behavior:
Claude Code should distinguish between "generating harmful content" and "processing existing local documentation for translation/formatting". Currently, the tool becomes unusable for security educators as the automated agents refuse to even list files in a security-related repository.

Steps to reproduce:

1. Create a folder named ch27-pedagogical-ransomware.

1. Ask Claude Code: "List all markdown files in this project to prepare for translation."

1. The agent blocks the request under the Cyber Policy.

Note: I have already submitted the "Cyber Use Case" exemption form, but the core issue lies in the agent's over-aggressive filtering of local file metadata.

Claude Model

Opus

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

Version: 2.1.92 Session name: traduction-french-to-english Session ID: d868adea-6f9b-416e-9073-0260bdf38bdc cwd: /home/vero/Documents/BAD Login method: Claude Max Account Organization: nicolas.deoux@gmail.com's Organization Email: nicolas.deoux@gmail.com Model: Default Opus 4.6 with 1M context · Most capable for complex work MCP servers: 2 need auth · /mcp Setting sources: User settings

Platform

Anthropic API

Operating System

Ubuntu/Debian Linux

Terminal/Shell

Other

Additional Information

Version: 2.1.92
Session name: traduction-french-to-english
Session ID: d868adea-6f9b-416e-9073-0260bdf38bdc
cwd: /home/vero/Documents/traduction
Login method: Claude Max Account
Organization: nicolas.deoux@gmail.com's Organization
Email: nicolas.deoux@gmail.com

Model: Default Opus 4.6 with 1M context · Most capable for complex work
MCP servers: 2 need auth · /mcp
Setting sources: User settings