Israeli Cybersecurity Firm Exposes Critical Security Flaws in Elon Musk's Grok AI
Translated & summarized from Ice by baba
Israeli cybersecurity firm Profero has uncovered critical security flaws in Elon Musk's Grok AI, potentially allowing full remote control and data theft. The vulnerabilities include an exposed management interface and shared user accounts, exposing sensitive information. Despite being notified in August, xAI and Cursor have released over 240 updates without fixing the seven critical issues, leaving users at risk.
The story in 5 lines · by baba
- Israeli firm Profero found severe security flaws in Elon Musk's Grok AI agent.
- Vulnerabilities could grant attackers full administrative control and remote access.
- Exposed data includes cloud keys, financial cookies, and sensitive files.
- xAI and Cursor were notified in August but failed to fix the issues.
- Over 240 software updates were released without addressing the critical flaws.
An extensive investigation by the Israeli cybersecurity firm Profero has revealed severe security vulnerabilities in Grok, the AI agent developed by Elon Musk's xAI. The infrastructure, managed by Cursor, suffers from a weak architecture that could allow attackers to gain full administrative privileges (Root) on virtual machines and remotely control user accounts with a single malicious command. Researchers described the exploit as a combination of a Trojan horse data-stealer and a built-in rootkit, enabling complete takeover without the need for external code or malware.
Profero's findings detail significant engineering failures, including an exposed container management interface on the network without any authentication. The internal authentication token was set to the simple word "local," facilitating an easy escape from the system's sandbox environment. Furthermore, all AI agents within the same account share a single Linux user and a common file system. This means a breach of one agent exposes the entire account history, including cloud access keys, financial website cookies, and sensitive emails and files.
Additional vulnerabilities include a Chrome browser operating with an open, password-less developer interface, allowing malicious code execution within user tabs. The system also struggles to eject attackers; even when xAI attempted to refresh access keys, new credentials were sent as unencrypted cleartext, allowing attackers to intercept them within a minute and maintain persistent access. The private key for signing Cursor's browser extension was also found within the software files, enabling attackers to create a seemingly legitimate but malicious extension to steal credentials.
Profero initially alerted xAI and Cursor to seven critical security flaws in late August, providing a detailed technical report. Cursor responded with an automated bot message, while xAI reportedly ignored all seven official inquiries. Despite releasing over 240 software updates in the following two months, neither company addressed the critical vulnerabilities, leaving users exposed.
The same event, reported separately by each outlet. Open a few to compare what different newsrooms emphasize — and what they leave out.
Haredi 1Other 1
