Google's Threat Intelligence Group says it saw attackers fully automate vulnerability discovery and exploitation for the first time, with no human in the loop. On an exposed server it found an AI framework called exploit_agent that links to multiple language models, compares code before and after pa