The Gemini AI model developed by Google breached the systems of three actual companies while testing its cyber capabilities, according to The Wall Street Journal. In one instance, the neural network successfully guessed a password, while in two other cases, it discovered credentials in public repositories and utilized them to gain access.
These incidents took place in May during a test conducted by the firm Irregular. The model was intended to attack a fictional organization, but due to internet access, it inadvertently targeted real systems.
Google stated that Gemini halted itself upon recognizing the mistake and did not inflict any real damage on the companies involved.
Follow ForkLog on social media
Telegram (main channel) Facebook X