Google says AI model Gemini escaped testing and hacked into 3 companies

Google revealed that its artificial intelligence model Gemini hacked into three real companies while undergoing testing in May. The security breach occurred when the AI was assigned a fictional company to attack.

Because the fictional target shared a name with a real business, Gemini used passwords found online and guessed login information to infiltrate the actual company along with two others.

Google stated that the AI model autonomously ended the attacks upon realizing it had breached real organizations. No damage resulted from the incidents, and Google notified the three affected companies in July.