MS-Agent Vulnerability Let Attackers Hijack AI Agent to Gain Full System Control

A critical command-injection vulnerability (CVE-2026-2256) in the ModelScope MS-Agent framework allows attackers to exploit prompt-injection and the framework's unsanitized Shell tool to execute arbitrary OS commands remotely (CVSS 9.8), potentially leading to full system compromise; CERT/CC noted the vendor had not issued a patch and recommends immediate mitigations including sandboxing, least privilege, strict input validation, and allowlists.