ChatGPT’s Mac app was storing chats in plain text
OpenAI just patched a glaring security flaw in its ChatGPT macOS app, which was previously storing user conversations in unencrypted plain text on local drives. That meant any malicious app with machine access could easily read your chat history. Following external reports, OpenAI quietly pushed an update that finally encrypts local conversation records.
If you installed OpenAI’s official ChatGPT app on your Mac recently, everything you discussed with it might have been sitting on your hard drive completely unencrypted.
Until an update dropped on Friday, the macOS app was quietly storing all your conversation records in plain text. Any malicious software running on your computer could read your entire chat history without breaking a sweat.
A simple rename exposed full chats
The flaw was first spotted by developer Pedro José Pereira Vieito, who posted his findings on Threads. To demonstrate the danger, he built a simple demo tool that pulled and displayed ChatGPT conversations in real time with the click of a button.
Reporters at The Verge confirmed the vulnerability firsthand. It did not take complex forensic tools: simply locating the local data files and changing their names revealed the raw transcripts.
After being contacted about the issue, OpenAI released an update designed to encrypt stored conversations. OpenAI spokesperson Taya Christianson stated that the company shipped a fix to encrypt these records while maintaining user experience and security standards. Tests with the updated version confirmed that plaintext access was blocked.
The cost of skipping Apple’s sandbox
How did a mistake this basic make it into production? It all comes down to distribution.
Pereira Vieito noticed the issue while investigating why OpenAI opted out of Apple’s App Sandbox protections. Because OpenAI chose to distribute the macOS client directly from its website rather than through the Mac App Store, it bypassed Apple’s mandatory sandboxing rules.
While users generally understand that OpenAI may review prompts for training unless opted out, nobody expected local transcripts to be accessible to arbitrary software running on the same machine.
Why it matters
We think this slip-up serves as a blunt reminder of what happens when AI labs prioritize shipping velocity over fundamental software security.
For everyday users and developers, local AI wrappers are trusted with sensitive company code, private thoughts, and confidential documents. If you have the ChatGPT macOS app installed, you need to check for updates immediately to make sure your past sessions are not left hanging in plaintext.
Distributing outside the Mac App Store gave OpenAI freedom from Apple’s review delays and system restrictions. But skipping the sandbox also means giving up the built-in safety net that prevents local files from being siphoned off. Storing user prompts in plaintext is an amateur mistake, and as AI moves deeper into our operating systems, users will likely demand much higher transparency about how local data is handled.
Sources (1)
- theverge.com(opens in a new tab)Originaltheverge.com