CVE-2026-14898 describes an information exposure vulnerability in the OpenAI Codex desktop app for macOS. The app renders remote images from Markdown in model responses. An attacker who can place an indirect prompt injection into content processed by Codex (e.g., from connected-tool results or other untrusted sources) can induce the model to generate a remote image URL containing sensitive data from the Codex session. The app then automatically fetches the remote image URL, causing the embedded sensitive information to be sent to an attacker-controlled server without requiring user interaction. This vulnerability enables exfiltration of secrets including API keys, source code, and connected tool data. No integrity or availability impacts were demonstrated, and no exploitation in the wild is known. No patch or official remediation guidance has been provided as of the published date.

Successful exploitation could result in unauthorized disclosure of sensitive information accessible within the Codex session, such as API keys, source code, and data returned by connected tools. There is no evidence of impact on data integrity or system availability. No known active exploitation has been reported.

Patch status is not yet confirmed — check the vendor advisory for current remediation guidance. Until an official fix is available, users should exercise caution when processing untrusted content in the Codex desktop app for macOS and avoid scenarios where indirect prompt injection might occur. Monitor vendor communications for updates on remediation.