OpenAI lately launched an official ChatGPT app for macOS, which can be the primary ChatGPT app for any desktop platform. Whereas having an app makes the method of speaking to the chatbot extra seamless, some customers have raised privateness issues. That’s as a result of the app shops all conversations in plain textual content, which might expose delicate consumer knowledge.
As famous by developer Pedro Vieito in a put up on Threads, the official ChatGPT app for Mac doesn’t use the usual macOS sandbox system. By wanting on the choice and cache recordsdata saved by the app, Vieito seen that each one conversations registered within the app are saved saved in plain textual content, making it straightforward for anybody to entry this knowledge.
However what precisely does all this imply and why isn’t the ChatGPT app sandboxed?
Privateness and sandboxed apps on macOS
For these unfamiliar, “sandboxing” is a safety management system that primarily runs an app and all its knowledge in an remoted atmosphere. This fashion, the app can’t entry different components of the system with out permission, simply as different apps can’t simply learn knowledge from a sandboxed app.
On iOS, all third-party apps run underneath a sandbox. However on the Mac, this technique was solely carried out with OS X Lion in 2011. Years later, with macOS Mojave, Apple added new layers of safety in order that apps all the time ask the consumer’s permission to entry knowledge outdoors their sandbox.
Nonetheless, whereas working an app in sandbox makes every thing safer, this technique stays non-compulsory on macOS as some extra complicated apps require full disk entry. And there are a variety of macOS apps that aren’t sandboxed. Nonetheless, in relation to chat apps dealing with delicate knowledge, most of them are sandboxed.
However what are the implications for customers?
Studying ChatGPT conversations with out consent is kind of straightforward
Anybody can discover the conversations from the ChatGPT app by going to Library > Utility Help > com.openai.chat. And because the conversations are saved outdoors a sandbox and in plain textual content, this additionally implies that the conversations will be accessed by different apps, processes, and even malware working on the Mac – all with out the consumer ever figuring out.
9to5Mac was capable of verify that ChatGPT conversations are saved in plain textual content. We additionally constructed a device to gather knowledge from the ChatGPT app with only a click on, and it really works with out asking for any permission.
In fact, OpenAI’s privateness insurance policies make it clear that each one your conversations with ChatGPT will be collected by the corporate to enhance its language mannequin. That alone is an efficient purpose by no means to share delicate knowledge with ChatGPT. However figuring out that this knowledge may find yourself in anybody’s fingers makes every thing worse.
We’d prefer to imagine that OpenAI simply made an oversight by not sandboxing the ChatGPT app. However for now, OpenAI has but to touch upon customers’ issues.
And in case you’re actually involved about privateness, a great way to maintain your knowledge protected on macOS is to solely obtain apps from the Mac App Retailer or verify whether or not an app downloaded from an exterior supply runs in sandbox.
Learn additionally
FTC: We use revenue incomes auto affiliate hyperlinks. Extra.
