Could Your Phone's AI Be Tricked? What Aussies Need To Know

You've probably heard of AI assistants like Google Gemini – the voice that helps you set alarms, send messages, or look up information on your phone. They're becoming more common, and for many of us, they're quite handy.

However, new research has highlighted a clever way these AI assistants could potentially be tricked. Imagine getting a notification from an app, say WhatsApp or Slack, which then subtly 'whispers' instructions to your phone's AI without you even realising. Researchers found that a malicious notification could essentially 'hijack' Google Gemini on Android phones.

The tricky part is something called 'indirect prompt injection'. Think of it like someone subtly changing a note that you then read aloud to your assistant, causing the assistant to do something unintended. In this case, the 'note' is a notification from another app. The researchers managed to make the AI perform actions like sending your private information or making calls without your direct knowledge, by hiding commands within these seemingly innocent notifications.

Google has been alerted to this vulnerability and is working on fixes. For now, it's a good reminder to be mindful of the apps you install and the permissions you grant them. As AI becomes more integrated into our everyday devices, understanding these kinds of security risks will become increasingly important for all of us.