Imagine for a moment that your trusted virtual assistant turns into a sneaky spy, collecting your every secret to whisper them to prying ears. Creepy, right? Yet this is the disaster scenario that almost happened with the ChatGPT app on macOS. But don’t panic, my friends—OpenAI patched the vulnerability before things got out of hand (spoiler alert, but I’m just looking out for your fragile nerves)!
It all began with the “Memory” feature in ChatGPT, designed to make the AI smarter by remembering your preferences, like a robot vacuum that knows you take your coffee with two sugars. Except in this case, a security researcher discovered that an attacker could inject malicious instructions into this memory.
The exploit, cleverly named “SpAIware” (kudos for the wordplay!), relies on a prompt injection attack from compromised websites or documents. By visiting a malicious site or analyzing an infected document with ChatGPT, the AI could store hidden instructions in its long-term memory.
The result? All your future conversations could be quietly sent to a server controlled by the attacker. Convenient for the bad guys, not so great for your privacy.
The scariest part? This vulnerability was persistent, like a stubborn computer virus. Even if you closed and reopened the application, the malicious instructions would remain active.
Fortunately, the engineers at OpenAI didn’t waste any time fixing the issue. Version 1.2024.247 of ChatGPT for macOS has now sealed this data exfiltration pathway.
The moral of the story: always keep your apps up to date, even if it’s a hassle, and don’t forget to regularly clear what ChatGPT has stored in its memory. You never know—it might still remember that time you unsuccessfully asked how to hack your ex’s Instagram account.
