Executive brief
Claude Code is an AI-assisted coding tool that runs in the terminal and executes code operations on behalf of users. A bug in its startup security dialog allowed the tool to execute commands from untrusted project files before prompting the user to approve the action, potentially enabling attackers to run arbitrary code by distributing malicious projects.
Technical details
This vulnerability is a code injection flaw (CWE-94) in Claude Code's startup trust dialog mechanism. The root cause is improper validation of code execution timing: the application executes commands from project files before enforcing the user-confirmation trust dialog, allowing code to run without explicit user consent. The attack vector is local/network-adjacent and requires a user to start Claude Code in a directory containing malicious project files; user interaction is required (the user must launch the application in an untrusted location). An attacker can achieve arbitrary command execution with the privileges of the Claude Code process and the invoking user. The vulnerability has been patched in version 1.0.111; users with auto-update enabled received the fix automatically.
Affected products
- Anthropic Claude Code < 1.0.111
Timeline
- 2025-10-03: disclosed: GHSA-4fgq-fpq9-mr3g advisory published