China-Linked AI Agents Exploit Zero-Day in MONAI Label Medical Imaging Software

Amazon Threat Intelligence detected a China-linked threat actor using autonomous AI agents to discover and exploit a zero-day vulnerability in MONAI Label, an open-source tool for AI-assisted annotation of medical images. The activity was identified in September after researchers found an exposed, unauthenticated instance of the Cairn AI penetration testing framework. The agents were tasked with obtaining shell access on university targets in East Asia outside mainland China and successfully mapped an application API before locating a path traversal flaw in a file upload function. Within roughly 11 minutes the agents moved from API mapping to arbitrary file writes, later planting a Python reverse shell that executed on interpreter startup to achieve root access. Amazon assessed the flaw as a zero-day with high confidence due to the absence of any prior advisory or fix in public databases. The MONAI Label developers subsequently released an advisory rating the issue critical for versions 0.8.5 and earlier, confirming that authentication is disabled by default and a patch will ship in the next release. Attribution to a Chinese-nexus actor was made with moderate confidence based on prompt language, script time zones, and the strategic nature of the targeted institutions.

SecurityWeekTranslated from English

Read at source

Amazon Threat Intelligence has disclosed that a China-linked threat actor deployed autonomous AI agents to identify and exploit a zero-day vulnerability in MONAI Label, an open-source framework used for AI-assisted annotation of medical images.

The activity came to light in September when researchers discovered an exposed instance of Cairn, an open-source AI penetration testing framework, that lacked any authentication. This allowed visibility into the operator’s prompts and the agents’ outputs, revealing a roughly 48-hour campaign targeting IP addresses belonging to universities in East Asia outside mainland China.

The operator supplied the agents with a list of targets and instructed them to obtain shell access. After identifying the software on one target and mapping its API, the agents located a path traversal vulnerability in a file upload function. “In about 11 minutes, the agents went from mapping the application’s API to arbitrarily writing files on the server,” Amazon stated.

In a subsequent session the agents leveraged the flaw to plant a Python reverse shell inside a file executed every time the Python interpreter starts, ultimately obtaining a root shell. Amazon assessed the vulnerability as a zero-day with high confidence because no corresponding fix, issue, advisory, or prior mention existed in vulnerability databases.

The operator later provided the agents with a proxy that tunneled into a second university network, where the same vulnerable software was already present, indicating prior compromise via the same zero-day. MONAI Label developers published an advisory this week rating the bug critical and affecting versions 0.8.5 and earlier. Authentication is disabled by default, allowing any network-reachable attacker to exploit the issue; a fix is scheduled for the next release.

Amazon attributed the operation to an unidentified Chinese-nexus threat actor with moderate confidence, citing the language used in prompts, the time zone reflected in scripts, and the likely strategic interest of the targeted universities to China.

Meridian digestThe week's most consequential security stories, distilled into one briefing.See the digest

More in APT & Nation-State

Armory Castle Exploits VMware vCenter Zero-Day CVE-2026-123890 to Obtain Admin Access and Exfiltrate Data
Chinese APT Armory Castle Exploits VMware vCenter Zero-Day CVE-2026-123890 for Over Two Months Before Patch
Chinese APT Groups Retool EggStreme Malware and Deploy New RATs Against Southeast Asian Governments