Security researchers just found a flaw in the ChatGPT Mac app that could hand over your entire chat history. Worse, it could let an attacker run commands while pretending to be you. The good news: it’s already patched, and keeping yourself safe takes about a minute.
What happened with the ChatGPT Mac app
Researchers at the Objective-See Foundation, a nonprofit that builds free Mac security tools, discovered the problem and reported it to OpenAI. A Wired report on the fix landed this week, and it’s worth five minutes of your time.
Here’s the short version. The bug let an attacker effectively take over ChatGPT on a victim’s Mac. That means access to all the chat logs and data stored by the app, plus its connections to things like browser sessions. The researcher behind the demo called the exploit “insanely trivial.” His proof of concept took about a dozen lines of code.
So yes, we’re talking about real access. Not a theoretical scare.
Before you panic, one clarification: this required local access. An attacker needed to run code on your machine (through malware, a malicious app, or hands-on access) to exploit it. That’s a smaller risk than a remote attack, but it matters, because plenty of Macs already carry malware their owners don’t know about.
What the flaw actually exposed
Let’s talk about why this specific bug stings.
Your ChatGPT history is more sensitive than most people realize. Scroll through yours. There are medical questions, work documents pasted in for rewriting, maybe financial details, maybe code from your employer. It’s a diary of what you needed help with.
A attacker with those logs learns a lot. And it gets worse: the flaw also let ChatGPT be used as a puppet. An attacker could make the app run commands, like opening a browser or touching other sensitive applications, and macOS would treat those actions as legitimate instructions from OpenAI’s software. Your security tools see a trusted app doing suspicious things. That’s a bad combination.
This is the second time the Mac app has drawn security attention. Back in 2024, an early version stored conversations in plain text, readable by any app on the machine. OpenAI fixed that too. Two incidents don’t make a pattern by themselves. But they do suggest the app’s deep system access deserves a bit of suspicion from its users.
How attackers get local access in the first place
Local attack sounds abstract, so here’s the plain-English version. Malware lands on Macs all the time through pirated apps, fake “free” utilities, sketchy browser extensions, and downloads pretending to be something they’re not. The moment one of those runs, whatever privileges you granted it are fair game. On a machine like that, the ChatGPT Mac app becomes a loot box: chat logs in one grab, plus a trusted channel for running commands that look legitimate.
This is also why the ChatGPT Mac app security problem shouldn’t be read as “OpenAI got hacked.” Nothing happened on OpenAI’s servers. The weakness was in how the desktop app handled its power on your machine, which is a different (and very fixable) kind of mistake.
The one thing to do right now
Update the app. Today.
Open ChatGPT on your Mac, click the app menu next to the Apple logo, and choose the update or “check for updates” option. The patched version is already rolling out. If you installed the app through the Mac App Store, updates happen automatically, but you can force one from the Updates tab.
Not sure which version you’re on? Check the app’s “About” screen and compare it with the current version on OpenAI’s official site. When in doubt, download a fresh copy from OpenAI directly. Never from a random download site, which is a classic way to get malware instead of an app.
That’s it. The flaw is patched in current builds, so updating closes the door. Everything below is about making sure the next one doesn’t catch you.
Why AI apps are juicy targets
It helps to understand why attackers bother with ChatGPT at all.
AI apps sit in a strange position on your computer. To be useful, they get deep access: they read your clipboard, talk to your browser, touch your files when you ask, and store long histories of your private questions. Most apps pick one or two of those privileges. AI assistants want all of them.
That makes the apps themselves a target. Breaking into one is like stealing a master key instead of picking a single lock. The same researcher who found this bug recently spotted a similar issue in the dictation feature of Meta’s new Muse assistant. He says he’s already filed a new report about the integration between ChatGPT and OpenAI’s always-on Dots assistant, which OpenAI is reviewing.
Expect more of these. AI apps are growing fast, and security review tends to trail behind new features. Every month brings another assistant that wants to live on your desktop, read your notifications, and sit inside your browser. More AI software on more machines means more chances for a flaw like this one. The trust these apps get from your operating system is exactly what makes them worth attacking. Our guide to local AI security fixes covers this pattern from another angle: the tools are only as safe as the setup around them.
Four habits that keep your AI apps safer
You don’t need to become a security nerd. These four habits cover most of the real risk.
1. Turn on auto-updates for every AI app. The ChatGPT flaw was patched quickly. That only protects people who actually got the update. Speed matters, and humans are bad at checking manually.
2. Never paste what you can’t afford to leak. Passwords, client data, unreleased financials, medical records. If it would hurt to see it in a screenshot online, it doesn’t belong in a chat window. Our list of files you should never upload to ChatGPT has the specifics.
3. Trim your history now and then. Delete old conversations you don’t need. Less stored data means a smaller prize if something goes wrong later.
4. Treat every AI app like it has a key to your house. Because on your Mac, it mostly does. Before installing the next shiny assistant, ask what access it actually needs, and whether you’d be okay with that access falling into the wrong hands someday. If you share sensitive work chats, our guide on protecting your ChatGPT history goes deeper.
None of these take more than a few minutes. Together, they shrink your exposure a lot.
The takeaway
The ChatGPT Mac app had a serious flaw, researchers found it, and OpenAI fixed it. Your move is simple: update the app today, and skim your history for anything you’d rather not have sitting there. The apps are getting more capable every month. A little skepticism, aimed at the software itself, is now part of using them well.
