Claude Code Flaws Could Let Malicious Repositories Run Commands and Steal API Keys

· Original article ↗

Summary

Check Point researchers found Claude Code configuration flaws that could trigger commands without approval, bypass MCP consent, and redirect API traffic to steal keys. Anthropic patched the issues and added warnings and confirmations.

Key points

  • Repository configuration files could include hooks that automatically run arbitrary commands when a project is initialized, without asking the user for approval.
  • MCP integration settings could override user approval for external actions, bypassing consent mechanisms.
  • Manipulated configuration could redirect Claude Code API traffic to an attacker-controlled server, potentially exposing API keys and credentials.
  • Attackers could target developers through a malicious repository they clone, or through malicious insiders or pull requests.
  • Check Point reported the vulnerabilities to Anthropic from July to October 2025; Anthropic rolled out fixes after the reports and added warnings and user confirmation for potentially dangerous actions.

Article Details

Vulnerability Types
  • Arbitrary command execution through project configuration hooks
  • MCP approval and consent bypass
  • API traffic redirection enabling API-key and credential exfiltration
Exploitation Status
unknown
Exploit Availability
unknown
Patch Status
available

MITRE ATT&CK

Vendors

Products

Related Articles