OpenAI patches critical ChatGPT Mac app vulnerability

OpenAI patched a critical security vulnerability in its ChatGPT macOS app on September 25. The flaw could have allowed attackers who already had malware on a victim's machine to access sensitive chat logs and execute arbitrary commands on the computer.

The vulnerability was insanely trivial to exploit. It required only about a dozen lines of code. Attackers could manipulate a trusted script interpreter component within the app to deliver and run malicious scripts. This method provided a pathway to significant system access.

Discovery and response

Researchers at the Objective-See Foundation discovered the security flaw. Patrick Wardle, a software analyst at the foundation and a longtime macOS researcher, was responsible for the identification.

OpenAI publicly acknowledged the flaw and the subsequent fix in its system change log on the September 25 date. A company spokesperson, Shane Bauer, stated that while they continue to evolve their security practices, they recognize a need to move faster. The issue was resolved following a responsible disclosure process where researchers privately reported the finding before any public announcement.

Impact and broader implications

The potential consequences of the vulnerability were severe. It could have allowed attackers to take over the ChatGPT application on a victim's computer. From there, they could access not only chat logs but also browser sessions and data from other sensitive applications.

This incident arrives as enterprise adoption of AI tools like ChatGPT has exploded over the past year. Millions of users now integrate these applications into daily workflows for tasks like drafting emails and analyzing confidential business documents. Patrick Wardle noted that AI companies are often more focused on adding new features than on security, which broadens the potential attack surface. Security researchers consistently warn that the race to market frequently leaves security as an afterthought in AI application development.

Responsible disclosure and ongoing research

The disclosure process for this flaw followed standard security research protocols. Wardle has also recently identified a patched flaw in Meta's Muse AI assistant dictation feature. Also, he has submitted a new vulnerability finding to OpenAI related to ChatGPT's integration with the always-on Dots AI assistant. OpenAI is currently reviewing that report.

Wardle will present a broader analysis of AI macOS application bugs at the Objective by the Sea security conference in November.