A critical security flaw in OpenAI Codex CLI and Desktop allows unauthenticated attackers to bypass user approval and execute arbitrary code by exploiting a command-safety parsing error.

This vulnerability involves the improper neutralization of escape sequences (CWE-150), where the command-safety parser fails to correctly interpret the PowerShell stop-parsing token. An unauthenticated attacker can supply a malicious repository that tricks the application into executing file-writing commands without user authorization, eventually leading to full remote code execution under the user's privileges.

Successful exploitation permits an attacker to read, modify, or delete files within the scope of the user's account, resulting in significant data compromise and loss of integrity. Given the CVSS score of 8.8, the risk is categorized as High, as it allows unauthorized code execution that could lead to full system compromise if the application is running with elevated privileges or has access to sensitive environments.

Immediate Action: Upgrade to OpenAI Codex CLI version 0.131.0 or later, or update Codex Desktop to the relevant fixed version (26.519.22136 for macOS, 26.519.21041 for Windows, or 26.519.2081.0 for the Microsoft Store package).

Proactive Monitoring: Review system logs for unauthorized file-writing activity or unexpected process spawning originating from the Codex application.

Compensating Controls: Ensure filesystem permissions are strictly enforced to minimize the impact of unauthorized writes, and avoid opening repositories from untrusted or unknown sources until the software is patched.

Public Exploit Available: No

The vulnerability presents a significant risk to user integrity and data security. Organizations utilizing OpenAI Codex tools should prioritize the deployment of the provided security updates across all affected environments immediately to eliminate the possibility of approval bypass and subsequent arbitrary code execution.