ChatGPT for Mac app logged queries in an unencrypted file before getting caught

Posted:
in Mac Software

Apple has strict guidelines about protecting user data with sandboxing, but ChatGPT for Mac bypassed all of this by storing conversations in plain text until it was patched on June 28.

ChatGPT for Mac showing a text field in place of Spotlight
ChatGPT for Mac



When everything is working the way it should on Mac, data should be siloed between apps so no single app can access another app's data without APIs or user permission. ChatGPT decided to ignore Apple's guidance and broke that structure by opting out of sandboxing and storing user conversations in plain text.

Storing files this way left them open for any other Mac app to find and read them freely. That means if a user's Mac was infected with malware or malicious apps, the private data shared with ChatGPT could be read freely.

Pereira Vieito discovered the problem and shared it on Threads.

Post by @pvieito
View on Threads



An update to ChatGPT for Mac was issued on Friday to patch this problem. All data from using ChatGPT is now hidden behind encryption.

"We are aware of this issue and have shipped a new version of the application which encrypts these conversations," OpenAI spokesperson Taya Christianson says in a statement to The Verge. "We're committed to providing a helpful user experience while maintaining our high security standards as our technology evolves."

When an app is submitted to the Mac App Store or for it notarization, it goes through a review process that ensures the app handles data via sandboxing. It is a method that ensures apps only have access to the data they have and none else on the system.

OpenAI's ChatGPT for Mac app is distributed from the web and doesn't use sandboxing. The app can access private data the user shares, like emails and confidential records, to perform whatever task the user asks.

If you've installed ChatGPT for Mac, ensure it has been updated to the latest version. While the vulnerability likely wasn't taken advantage of in the short time since the app launched, it is still a silly mistake for a company like OpenAI to make.

The ChatGPT for Mac app is separate from the larger partnership OpenAI has with Apple. Later in the fall, users can opt to send some requests to ChatGPT instead of Apple Intelligence as a part of macOS Sequoia.



Read on AppleInsider

Comments

  • Reply 1 of 20
    sflocalsflocal Posts: 6,107member

    "We are aware of this issue and have shipped a new version of the application which encrypts these conversations," OpenAI spokesperson Taya Christianson says in a statement to The Verge. "We're committed to providing a helpful user experience while maintaining our high security standards as our technology evolves."


    This is such a pure BS response from them.  "As technology evolves"?  Bullsh!t.  Some jackass over there made the conscious decision to do this, or out of sheer coding laziness.  There ought to be more consequence than just a typical "lol... sorry, my bad" on ChatGPT's part.  How's about banning them from MacOS for a year?  Something?
    gregoriusmbaconstangAnilu_777StrangeDaysAlex1N9secondkox2dewme
  • Reply 2 of 20
    DAalsethDAalseth Posts: 2,846member
    OpenAI being sloppy with user data? I’m shocked, SHOCKED I tell you. /s
    timpetusTomPMRIbaconstangrobin huberAnilu_777StrangeDaysAlex1N9secondkox2
  • Reply 3 of 20
    Er, isn’t the author conflating sandboxing and encryption? 
    gatorguywilliamlondon9secondkox2
  • Reply 4 of 20
    baconstangbaconstang Posts: 1,122member
    This displays the difference between saying you're about privacy, and actually being about privacy.
    williamlondonAnilu_777Alex1N9secondkox2
  • Reply 5 of 20
    robin huberrobin huber Posts: 3,989member
    This is what beta testing is all about. Never downloaded a public beta and never will. 
    williamlondonbaconstangdanox
  • Reply 6 of 20
    Anilu_777Anilu_777 Posts: 557member
    OpenAI just didn’t sandbox their app??? Who the f is coding it and who will take full responsibility?? Glad I’ve got a 14 Pro Max and won’t deal with this crap. 
    Alex1N
  • Reply 7 of 20
    jfabula1jfabula1 Posts: 139member
    Hmmmm, Elon is right about open AI???. Be careful out there.
    williamlondon9secondkox2
  • Reply 8 of 20
    StrangeDaysStrangeDays Posts: 12,949member
    jfabula1 said:
    Hmmmm, Elon is right about open AI???. Be careful out there.
    Incorrect. His crockpot crank tinfoil hat nonsense had nothing to do with this, which was simply sloppy amateur coding. 
    williamlondonAlex1N9secondkox2dewme
  • Reply 9 of 20
    danoxdanox Posts: 3,090member
    sflocal said:

    "We are aware of this issue and have shipped a new version of the application which encrypts these conversations," OpenAI spokesperson Taya Christianson says in a statement to The Verge. "We're committed to providing a helpful user experience while maintaining our high security standards as our technology evolves."


    This is such a pure BS response from them.  "As technology evolves"?  Bullsh!t.  Some jackass over there made the conscious decision to do this, or out of sheer coding laziness.  There ought to be more consequence than just a typical "lol... sorry, my bad" on ChatGPT's part.  How's about banning them from MacOS for a year?  Something?
    OpenAI probably isn't good at coding outside of OpenAI itself, in short making anything that works at OS level that's Apples job and Apple can't assume OpenAI knows how.
    edited July 4 Alex1N
  • Reply 10 of 20
    danoxdanox Posts: 3,090member

    jfabula1 said:
    Hmmmm, Elon is right about open AI???. Be careful out there.
    The A__ (Narcissist) is just jealous that his time is up.
    williamlondon9secondkox2
  • Reply 11 of 20
    dutchlorddutchlord Posts: 240member
    That’s why don’t want any “AI service provider” on my device. 
    williamlondonDAalsethAlex1N9secondkox2
  • Reply 12 of 20
    CaptainQCaptainQ Posts: 2member
    And we are still supposed to trust Apple and OpenAi for Apple Intelligence? 
    That's all really amateur and unprofessional.... The timing couldn't be worst 
    williamlondon9secondkox2
  • Reply 13 of 20
    coolfactorcoolfactor Posts: 2,290member
    It's right there in the name, people.

    "Open"AI.

    Open for all to see.

    But on a serious note, doesn't this reflect on Apple and macOS, too?

    Maybe the operating system should be more openly surfacing where an app is writing data to? The Installer does a fine job of that, but maybe make that a standard feature for running apps. too?
    edited July 4 williamlondonAlex1N9secondkox2
  • Reply 14 of 20
    tundraboytundraboy Posts: 1,901member
    See, you just can't trust these people ... 
    williamlondon9secondkox2
  • Reply 15 of 20
    Wesley HilliardWesley Hilliard Posts: 216member, administrator, moderator, editor
    Er, isn’t the author conflating sandboxing and encryption? 
    Nope
    jfabula1 said:
    Hmmmm, Elon is right about open AI???. Be careful out there.
    Nope
    Anilu_777 said:
    OpenAI just didn’t sandbox their app??? Who the f is coding it and who will take full responsibility?? Glad I’ve got a 14 Pro Max and won’t deal with this crap. 
    That's not how any of this works. The ChatGPT app on macOS has nothing to do with the ChatGPT handoff that occurs thanks to the company's partnership. Apple Intelligence is made by Apple, powered by Apple Silicon on device or in an Apple-run server. If a user sends a query to ChatGPT from their device, it is with explicit permission and through a separate service channel that doesn't log IP, the query, or any attached data.

    BTW, encryption and sandboxing are very separate things but can work together. When ChatGPT updated their app, note that the data storage location didn't change. ChatGPT didn't opt into sandboxing. Instead, it chose to encrypt the data in place. All sandboxing is is a permission structure where data is siloed into different places associated with the app that created it and you have to have permission to hand data between silos. ChatGPT didn't and still doesn't silo its data. It's just there open for anyone to get, but now it is at least encrypted.
    magnuskrantzdanoxwilliamlondon
  • Reply 16 of 20
    9secondkox29secondkox2 Posts: 2,896member
    I said it before and it beats repeating - keep chatgpt and anything from open ai the heck away from my devices. 
    edited July 4
  • Reply 17 of 20
    9secondkox29secondkox2 Posts: 2,896member
    jfabula1 said:
    Hmmmm, Elon is right about open AI???. Be careful out there.
    He’s right about a lot of things. Not everything - the guy is human, just like Jobs. But he’s a genius and has done a lot in terms of making the world a better place. 

    Haters gonna hate though. 
  • Reply 18 of 20
    Er, isn’t the author conflating sandboxing and encryption? 
    Nope
    The author may not be conflating the two but the way the article is written it certainly reads like he thinks that sandboxing and encryption are the same thing. 
    williamlondon
  • Reply 19 of 20

    jfabula1 said:
    Hmmmm, Elon is right about open AI???. Be careful out there.
    He’s right about a lot of things. Not everything - the guy is human, just like Jobs. But he’s a genius and has done a lot in terms of making the world a better place. 

    Haters gonna hate though. 
    You being Musk fanboy explains how you regular get things completely wrong  but are completely confident you are correct. 
    muthuk_vanalingamdanoxwilliamlondon
  • Reply 20 of 20
    dewmedewme Posts: 5,523member
    sflocal said:

    "We are aware of this issue and have shipped a new version of the application which encrypts these conversations," OpenAI spokesperson Taya Christianson says in a statement to The Verge. "We're committed to providing a helpful user experience while maintaining our high security standards as our technology evolves."


    This is such a pure BS response from them.  "As technology evolves"?  Bullsh!t.  Some jackass over there made the conscious decision to do this, or out of sheer coding laziness.  There ought to be more consequence than just a typical "lol... sorry, my bad" on ChatGPT's part.  How's about banning them from MacOS for a year?  Something?
    Of course it is BS. Unfortunately it is also entrenched as the "standard" response that corporate spokespeople say whenever they get caught with their pants down. I don't know who initially drafted the template for this and all of the other fully expected canned response, but he or she should at the very least get attribution for the reuse of this standard response.

    It goes something like this ...

    "We at _<insert corp name>_ take _<insert the name of what you just screwed up>_ very seriously. We are fully committed to ensuring that our customers are our top priority when it comes to _<insert the name of what you just screwed up>_ in all of our current and future products."

    Of course there are variations on the theme. But the gist of it is basically the same, gloss over what just happened and then state that the entire company has always been fully committed to ensuring what just happened could never happen. But it just did - and they don't want to talk about specifics any more. Let's just talk about visions, goals, and happy unicorns.

    Welcome to the bullshitosphere. 
    muthuk_vanalingam
Sign In or Register to comment.