AI Agents Exploited MONAI Label Zero-Day, Amazon Says
Amazon says AI agents linked to a China-connected threat actor exploited a MONAI Label zero-day and gained root access to university servers in East Asia.
Amazon’s threat intelligence team detected a roughly 48-hour campaign in September against IP addresses assigned to universities in East Asia outside mainland China. The company attributed the activity with moderate confidence to an unidentified China-linked threat actor.
Investigators found the campaign after locating an exposed instance of Cairn, an open-source framework for AI-assisted penetration testing. The instance had no authentication, allowing them to view the operator’s instructions and the agents’ responses.
The operator gave the agents a list of targets and instructed them to obtain shell access, which allows commands to run on a server. After finding MONAI Label, an open-source tool for annotating medical images, on one target, the agents mapped its application programming interface.
The agents then found a path traversal flaw in a file-upload function. The vulnerability allowed files to be written to arbitrary locations on the server. Amazon reported that the agents went from mapping the application’s API to writing files in about 11 minutes.
In a later session, the agents used the flaw to place a Python reverse shell in a file that runs whenever the Python interpreter starts. The action gave them a root shell, the highest level of access on the affected system.
Amazon assessed with high confidence that the flaw was a zero-day because no fix, security advisory or earlier vulnerability database entry existed for the project when the campaign took place. A zero-day is a vulnerability exploited before developers release a patch.
The operator later supplied a proxy connection to a second university and instructed the agents to target other devices on its network. Amazon found that the proxy server ran the same vulnerable software and inferred that the actor had used the flaw to compromise the server and create a tunnel into the university’s network.
MONAI Label developers published a security advisory this week rating the vulnerability critical. It affects versions 0.8.5 and earlier. Authentication is disabled by default, so anyone who can reach a MONAI Label server over a network could attempt to exploit the flaw. A fix has been created and is scheduled for the next release.
Amazon based its assessment of the actor’s China link on the language in the prompts, the time zone recorded in scripts and the likely strategic interest of the targeted universities to China.







