OpenAI patched ChatGPT macOS flaw CVE-2026-100754

OpenAI fixed CVE-2026-100754 in ChatGPT for macOS on September 25. The company says version 26.924.20706 contains the fix for a vulnerability that Patrick Wardle found could let malware already running on a Mac reach sensitive data and issue commands through trusted app components.

OpenAI’s changelog credits Wardle, a researcher with the Objective-See Foundation, with reporting the vulnerability. The note confirms the CVE, fix and version, but does not describe the root cause, severity or affected pre-fix releases.

According to secondary reporting on Wardle’s research, exploitation required malware to be installed on the target Mac; the proof of concept did not describe a remote compromise. In that account, ChatGPT components authenticated requests with digital-signature checks and also examined the requesting process’s parent and grandparent. The proof of concept passed an untrusted script to a trusted script interpreter, nesting that interpreter three times to satisfy the process-ancestry checks before reaching the main ChatGPT process.

The reported potential impact included access to ChatGPT chat logs, other data stored by the app and connected resources such as browser sessions. The flaw could also have allowed an attacker to direct ChatGPT to run commands involving a browser or other sensitive applications, while making the requests appear to come from OpenAI software. This local-app vulnerability is distinct from a separate disclosed breach chain involving OpenAI employee accounts.

The available sources do not establish that CVE-2026-100754 was exploited in the wild, that user data was accessed or how many installations were exposed.