In a somewhat chilling revelation, AI agents were able to find and exploit known vulnerabilities, but only under certain conditions, which researchers say indicates they're not close to being a ...