With the release of its ChatGPT desktop app for Mac last month, OpenAI provided chatbot interaction capabilities such as text generating and code writing. According to a recent study, there are security worries as the app may have had a weakness that allowed hackers or bad applications to access users' chatbot interactions. This app problem was addressed by OpenAI for ChatGPT.

We have released an updated version of the program that encrypts these discussions because we are aware of this problem. As our technology advances, we're dedicated to delivering a beneficial user experience while upholding our strict security requirements, Taya Christianson, an OpenAI representative, told The Verge.

What is the purpose of the ChatGPT macOS bug?

When Pedro Vieito, a developer, discovered the flaw, he revealed that all of the user's chatbot interactions were being kept in plain text on the computer. This implies that the talks may have been readily accessed by a hacker or malicious program that gained access to the machine.

This was risky as the application may have "skipped" Apple's built-in security control mechanism known as "sandboxing." A security feature called sandboxing keeps programs and their data separate, preventing unauthorized access to other system data or data from other apps. In essence, this guards against viruses or other programs accessing user data without authorization.

On macOS, programs must specifically request permission to access data outside of their sandbox, in contrast to iPhones, where all apps are sandboxed. Interestingly, sandboxing is usually used by chat applications that handle sensitive data to provide an additional degree of protection. It appears that the ChatGPT software was not sandboxed.
How to protect oneself

Firstly, make sure you only download programs from the official app store since they go through a screening process before being made accessible to consumers. When interacting with chatbots, avoid disclosing private information since it might be utilized to train artificial intelligence models.