OpenAI Codex Command Injection Vulnerability Let Attackers Steal GitHub User Access Tokens

**Executive Summary:** BeyondTrust researchers discovered a critical command-injection vulnerability in OpenAI Codex that permitted attackers to inject shell commands via GitHub branch names and steal GitHub OAuth/installation tokens (including from local auth files), enabling lateral movement and widespread compromise through automated malicious branches; the flaw affected web, CLI, SDK, and IDE integrations and was patched after responsible disclosure in January 2026.