Researchers from Tel Aviv University demonstrated a novel AI attack, dubbed "promptware," exploiting Google's Gemini AI. By embedding malicious instructions within seemingly innocuous calendar appointments, they successfully manipulated Gemini into controlling Google smart home devices. When Gemini summarized the calendar, it executed the hidden commands, turning on appliances like boilers. This marks a significant development, representing the first real-world impact of a prompt injection attack on smart home devices.

Prepared by Jonathan Pierce and reviewed by editorial team.