Critical Argument Injection Vulnerability in Popular AI Agents Let Attackers Execute Remote Code Cyber Security News