Attackers’ AI Agents Found a Zero-Day in Medical-Imaging Software and Gained Root Access
AI-summarised brief · reviewed before publication
Amazon Threat Intelligence reported that suspected China‑linked hackers deployed autonomous AI agents to exploit a zero‑day path‑traversal flaw in MONAI Label, an open‑source medical‑imaging server used by research teams. The vulnerability, present in version 0.8.5 and earlier, lets unauthenticated users write files outside the designated storage folder, eventually allowing the attackers to gain root access on a university server. The agents, running the open‑source Cairn framework, mapped the application in minutes and inserted a malicious Python startup file that was executed when a training process launched. Because MONAI Label runs with authentication disabled by default, any network‑reachable client could trigger the write, and investigators observed a intrusion attempt on a second university network using the same technique. The maintainers issued an advisory urging users to restrict access and monitor for file writes. A fix has been merged into the source code but no patched binary is yet released. The flaw is catalogued as CVE‑2026‑133968 with a critical 9.8 severity rating.
💡 Why It Matters
- · The incident proves AI‑driven agents can autonomously locate and weaponize zero‑day bugs, turning open‑source research tools into footholds for state‑linked espionage.