Profero Exposes Critical Security Flaws in xAI Grok and Cursor Platform
Israeli cybersecurity firm Profero uncovered critical security flaws in xAI's Grok agent and Cursor platform, allowing attackers to gain root access and steal sensitive data.
A comprehensive security report published by the Israeli cybersecurity firm Profero, led by founder Omri Segev Moyal, reveals deep architectural flaws in the core infrastructure of Elon Musk's xAI Grok AI agent. The platform, operated by Cursor, suffers from severe design vulnerabilities that allow an attacker to gain root administrative privileges and remotely control user accounts using a single malicious command.
According to the researchers, the flaw behaves as a lethal combination of a data-stealing Trojan and a built-in rootkit, enabling complete system takeover without requiring the injection of external malware. The report details embarrassing engineering oversights that allowed a full bypass of security controls, including an entirely unauthenticated container management interface left exposed on the network with the internal validation token hardcoded as "local".
Furthermore, all AI agents under a single account share a common Linux user and file system. Consequently, compromising one agent exposes entire account histories, including AWS S3 cloud access keys, cookies from financial accounts, and unrestricted access to emails and sensitive files. The built-in Chrome browser also operated by default with an open, passwordless debugging interface.
Despite xAI and Cursor releasing over 240 software updates following Profero's initial disclosure in late August, none of the seven critical security vulnerabilities were patched, leaving users vulnerable to ongoing remote exploitation.
