Vulnerability in OpenAI Coding Agent Could Facilitate Attacks on Developers

SecurityWeek reports that OpenAI Codex CLI contains a vulnerability, tracked as CVE-2025-61260, which can be exploited for command execution and was patched with Codex CLI 0.23.0 after being reported to OpenAI in August. The issue arises because Codex CLI automatically loads and executes commands defined in local project configurations, which are implicitly trusted and run without user approval, enabling an attacker who can commit or merge a crafted configuration file to plant malicious commands.

According to Check Point, an attacker could deploy a reverse shell, silently execute arbitrary commands, exfiltrate credentials and other secrets, escalate privileges, and move laterally, with the potential for supply-chain attacks if CI or build agents run Codex on checked-out code. The article notes that compromised templates or popular open-source projects could weaponise downstream consumers with a single commit. Written by Eduard Kovacs, the piece appeared on 2 December 2025.