Published on: August 07, 2025 12:33 pm IST
Imagine that AI is being used to trigger action in your smart home. This is what researchers have performed using Google Mithun AI Bot.
Imagine what hackers can use a popular AI chatbot, such as Google Gemini, to manipulate your physical environment, such as closing lights and equipment, and potentially large, triggering more dangerous events below the line? A science-fis seem out directly out of the film, but it is fine what researchers have performed by infecting a Google calendar invitation, which in turn allows them to kidnap Gemini and manipulate the real-world environment.
As Moldy By Wired, three security researchers performed by kidnapping GOOGLE’s primary AI assistant, Gemini, which is found on various Android phones. Researchers first achieved this by infecting a Google calendar invitation with instructions to change the status of electronic devices in a house. Later, when he asked Mithun to summarize the calendar invitations for the upcoming week, these infected instructions became active, eventually shut down the lights.
Researchers call it the first type of hack
This is reportedly the first time a generic AI system was infected to manipulate the actual world environment. This displays the risks that can pose large language models (LLM) as they are rapidly connected to physical objects, such as smart home devices and AI agents to complete integrated tasks.
It is part of comprehensive research, titled ‘Invitation is All You Need: Tara for Targeted Promptware Attack for Gemini-Interested Assistants.
“LLM is being integrated into physical humanoids, semi-and fully autonomous cars, and we really need to understand how to secure LLM, before we integrate them with this type of machines, where the results in some cases will not be safety and privacy,” Ben Nasi, one of the university’s researchers.
Google is taking action
The report also explained how Google knows about it. Google’s Andy Wayne claims that the weaknesses have not been exploited by hackers, but the company is taking them seriously. The report said the researchers behind the ‘invitation all need’ reached Google in February, and teams have been working on flaws since then and AI has been developing defensive mechanisms against quick injection attacks.
Mobile finder: iPhone 16 latest glasses and more
