“Google has developed a framework called Camel that aims to improve AI agent security by dynamically restricting an agent's possible actions based on the user's specific request.”