The Hacker News
AI Agents Uncover MONAI Label Zero Day and Gain Root Access
Suspected China-linked hackers employed autonomous AI agents to discover a previously unknown vulnerability in MONAI Label, an open‑source medical imaging server, and gain root access to a university system. The flaw, a path‑traversal weakness in components that store uploaded images and labels, allows an unauthenticated attacker to write files outside the server’s storage folder by appending a caller‑supplied file identifier to the storage path.
MONAI Label 0.8.5 and earlier are affected when deployed with the default, no‑login configuration. The maintainers have merged a fix in the source code, but no patched binary release exists yet to install. Amazon Threat Intelligence attributes the intrusion to the operator’s use of Cairn, an open‑source framework that runs AI agents as automated penetration testers; the Cairn control panel was found exposed on the internet without a password, revealing agent prompts, prompts history and task settings.
Investigators tie the campaign to a China‑linked actor with moderate confidence, citing Simplified Chinese prompts and time/target cues; however, no named group or government link has been established.
Amazon’s incident reconstruction describes the sequence: within roughly 11 minutes over a single session, the agents mapped MONAI Label’s interface and performed a write operation to the server, then later attained a root shell in another session via a startup‑time Python process that executed a planted file. A separate session allegedly provided a tunnel into a second university running identical software.
The October fix implements uploads that resolve inside approved folders and rejects empty names or names beginning with a dot; the official advisory notes no patched version is yet available and a fix will appear in a future release. CVE‑2026‑133968, rated 9.8/10, records the flaw, though the advisory page lacks a CVE entry at the time of reporting.
Practical steps advised include restricting access with firewall rules, enabling authentication, running MONAI Label under a limited account, and verifying that a server’s build contains the October fix before treating it as patched. The campaign’s impact claims acknowledge root access on one university server but stop short of confirming patient data exfiltration; variations in deployment may yield different outcomes.