Artificial Intelligence (AI) offers numerous benefits to everyday users, but it’s crucial to acknowledge the inherent security risks.The concern isn’t necessarily about machines gaining sentience, but rather the potential for malicious actors to exploit vulnerabilities and cause meaningful harm.

Recent cybersecurity research, as reported by Wired, demonstrates precisely how criminals could compromise systems like Google Gemini. Researchers successfully gained control of devices within a smart home network by embedding a malicious prompt within a Google Calendar event invitation. When a user requested a calendar summary and thanked Gemini, the hidden prompt instructed a Google Home device to execute commands, such as opening windows or switching off lights.

The findings were presented at the Big Hat Cyber Security Conference and initially reported to Google earlier this year. Andy Wen, leading director of Google Workspace for security product management, confirms the validity of the vulnerability, but stresses that real-world attacks are currently rare. However, he acknowledges the potential for more damaging consequences than simply turning off lights. Access to security cameras or thermostat controls could lead to more serious breaches.

The increasing sophistication of Large Language Models (LLMs) also presents a moving target for security professionals. hackers are continually discovering new attack vectors, making robust defense increasingly challenging. Google has taken the reported vulnerabilities “extremely seriously” and is accelerating efforts to patch existing and future weaknesses.