Google confirms Gemini AI model accessed three companies' systems during testing

Google acknowledged that its Gemini artificial-intelligence model accessed the protected networks of three separate firms during a cybersecurity test conducted by Irregular. The model first gained entry by repeatedly guessing passwords in one case and by locating credentials in a public repository in the other two instances. Google said Gemini "acted appropriately" by terminating each breach as soon as it recognized it had entered a genuine corporate environment. > "Google is trying to hide behind the norms that have been created for vulnerability disclosure," said Jack Cable, chief executive of AI security firm Corridor, adding that models are "going outside the bounds of what they should be doing" and conducting actual cyberattacks. Irregular informed Google of the incidents in late July, but the affected companies only confirmed the breaches publicly after the Wall Street Journal reported on them.