Security researchers expose flaw in ChatGPT's Mac app that could expose user data

OpenAI recently patched a critical security flaw in its ChatGPT macOS app that researchers say could have allowed attackers to seize control of the application and steal sensitive user information. The discovery underscores a growing concern as AI tools become more deeply integrated into everyday computing: the systems that require the most access are often the most attractive targets for compromise.

How the vulnerability worked#

Researchers at the Objective-See Foundation identified a weakness in how ChatGPT's components communicate with one another on macOS. The app is designed to verify that different processes are legitimate OpenAI software through digital signature checks—a safety mechanism built in across multiple layers to prevent malicious code from impersonating trustworthy components.

However, the team found a fatal gap: a trusted script interpreter that the app relies on would execute untrusted commands. By launching this interpreter multiple times in sequence, an attacker could bypass the layered verification system. "The vulnerability was insanely trivial to exploit," says Patrick Wardle, software analyst at Objective-See Foundation, noting that a working proof of concept required only about a dozen lines of code.

What attackers could access#

Once inside, a bad actor could potentially extract all of a user's ChatGPT conversations and data stored locally on the machine. Beyond that, the compromised app could be weaponized to execute commands on the victim's computer—controlling browsers, accessing other applications, and making requests that would appear to originate from legitimate ChatGPT operations.

A pattern across AI tools#

This incident reflects a wider tension in the AI sector. As Wardle explains, intelligent agents require extensive system access to function effectively. "They are like the building manager who has access to the keys to all the rooms," he says. When such privileged software is compromised, attackers gain a golden key to sensitive data and operations.

The researcher has uncovered similar flaws in other AI applications. Earlier, he identified a flaw in Meta's new Muse AI assistant that could have exposed authentication tokens. He has also recently reported a separate vulnerability to OpenAI concerning the integration between ChatGPT and the company's new always-on Dots AI agents—a report currently under review.

Industry prioritizes features over security