Hackers Hijacked Google’s Gemini AI With a Poisoned Calendar Invite to Take Over a Smart Home
1 min read
Summary
Researchers in Israel have demonstrated a series of attacks against Google’s flagship AI chatbot, Gemini, which demonstrates the potential for AI to be physically harmful in the real world.
Three researchers revealed the attacks at the Black Hat cybersecurity conference in Las Vegas, which show how AI can be used to carry out crimes such as cyber-spying and vandalism.
The attacks mostly start with a poisoned Google Calendar invitation, which includes instructions for the AI to carry out the crimes at a later time.
Google has introduced fixes for the vulnerabilities and said it took the issue “extremely seriously”
The research has “accelerated” the rollout of further defences against AI-prompt injection attacks, including more user confirmations when sensitive actions are undertaken by AI and machine learning to detect potential attacks.
