The light that lights unexpectedly, the enclosure that is launched Remove it from And the washer that is activated. No, the house is not possessed by a malicious spirit. But she may have been pirate.
In fact, as an investigation was revealed by the US media, three cyber security researchers have managed to take control of a remote house connected using the AI Gemini.
For this, these ethical pirates have sent an invitation from the Google agenda, which actually contains hidden instructions. Then, ask the AI assistant to summarize the next events during the week to activate the malicious program and the intelligent devices are lit.
Security problems
“The LLM (Great Language Model, Editor’s Note) is about to integrate into physical humanoids, in semi -autonomous and completely autonomous cars, and we must understand how to ensure the LLM before integrating them into this type of machines, where in some cases, the challenges will be security and non -confidentiality”, Ben Nassi alerts, a researcher at the University of Tel Aviv, American media.
But experts did not stop there. They discovered 14 different types of attacks against Gemini AI through an invitation from the Google agenda. During a demonstration at the Black Hat conference on Cybersecury in Las Vegas, researchers showed how Gemini could be used to send spam links, generate vulgar content, open the Zoom application and release a call, steal emails and gather details of a web browser.
The exploitation of these vulnerabilities does not require any particular knowledge, it reminds of ethical computer pirates. Developing such an attack would even be “accessible to all.” It is also possible to vary the techniques, using Gmail or GDOC instead of the Google agenda.
Google was informed of these security defects last February. Andy Wen, director of the Division of Security Products for Google Workspace, said that computer pirates have not used vulnerabilities with questionable intentions. But the company has implemented several corrections.
Source: BFM TV
