At the recent Black Hat cybersecurity conference in Las Vegas, researchers from Tel Aviv University unveiled a concerning vulnerability in Google’s Gemini AI, showcased through a technique called “Invitation Is All You Need.” This indirect prompt injection attack allows malicious Google Calendar invites to execute commands on smart home devices. By embedding harmful instructions into calendar events, the researchers revealed that simple responses like “thanks” could trigger actions such as turning on lights or heating systems. This marks a significant advancement in the capabilities of generative AI, enabling it to cause physical consequences. The team identified 14 unique exploitation methods involving actions like sending spam and deleting appointments. Due to the attack’s stealthy nature, users may struggle to trace malicious activities back to the original calendar invites. In response, Google is implementing enhanced security measures, including stronger detection and user confirmations for high-risk actions, recognizing the growing threats associated with AI integration.
Source link
