Google's Gemini AI accidentally hacks three companies during security testing

Google's Gemini AI model inadvertently accessed and hacked three real companies during security testing. The incidents occurred in May when a flaw allowed the AI to gain internet access, bypassing its test environment.