OpenAI Patches ChatGPT Mac App Flaw That Could Expose User Data
publishers 36articles 45first reported 28 Sep, 17:00 UTCdeveloping since 28 Sep · 32 editions
OpenAI has patched a vulnerability in the macOS version of its ChatGPT app that could have allowed attackers to access users’ conversations and other sensitive data, according to Wired.
Researchers at the Objective-See Foundation found that a trusted script interpreter within the app could accept an untrusted script and pass it into the main ChatGPT process. Although the app checked the digital signatures of communicating components and their parent processes, researchers said malicious code could satisfy those checks by spawning the interpreter three times.
Exploitation required an attacker to have already installed malware on the target Mac. From there, the flaw could potentially provide access to ChatGPT logs and connected resources, including browser sessions and other sensitive applications. It could also enable commands to appear as legitimate requests from OpenAI software. Objective-See researcher Patrick Wardle described the exploit as trivial and said his proof of concept used about a dozen lines of code.
OpenAI acknowledged the vulnerability and its fix in a September 25 system change log. Company spokesperson Shane Bauer told Wired that OpenAI was continuing to develop its security practices but recognized “a need to move faster.”
HOW THIS STORY WAS MADE
Written from 8 articles, 2 with the publisher's own text; 8 independent newsrooms once syndicated copies count as one; this version written 68 h after the record first saw the story; the editor kept it as written.
- Publishers
- 36
- Source articles
- 45
- Given to the writer
- 8, 2 with the article's own text
- Independent newsrooms
- 8
- This version written
- 68 h after the record first saw the story
- Second model (editor)
- kept as written
- Publication gate
- passed
- Human review
- none
WHAT THE EDITOR DID
The second model checked the writer's draft against the same sources and published it as written.
Written by a language model from the sources above, then checked by a second model that may only cut, attribute or correct. How it works →