Researchers Control Smart Homes Using Gemini Prompts

• Malicious prompts to Gemini can control smart home devices.
• The "promptware" technique utilizes a calendar for transmitting malicious commands.
• Google has implemented measures to improve the security of AI systems.

This is reported by Ars Technica, CNET.

Research presented at the Black Hat conference shows how malicious commands in the form of prompts for AI Gemini can control smart home devices.

Researchers from Tel Aviv University found that attackers can manipulate Google Home devices, such as lighting, heaters, and other elements of smart homes. Researchers demonstrated how, through calendar invitations, Gemini can be given commands that it executes, responding to common user requests.

Attackers utilize a technique called "promptware," where malicious instructions are embedded in the descriptions of calendar events. When a user queries Gemini about their schedule, the AI processes these malicious instructions, carrying out unintended actions. This method bypasses existing security systems of Google, exposing detected attacks.

Researchers believe this is the first incident where an attack on AI has real-world consequences. Furthermore, they indicated that through this technique, it is possible not only to control devices but also to send spam, delete events in the calendar, and open malicious websites.

Google has already implemented measures to address these vulnerabilities, including new checks on malicious instructions in invitations and documents. According to company representatives, their reaction to the research includes accelerating the introduction of new security methods.

These incidents highlight the growing risks associated with the integration of AI into everyday life. As AI systems become more powerful, attackers can exploit them for manipulation, putting user security at risk.

Tags: Technology/AI