Amazon’s threat intelligence division recently reported the detection of a zero-day vulnerability exploitation by autonomous AI agents linked to a Chinese threat actor. These agents targeted MONAI Label, an open-source tool used for AI-assisted medical image annotation.
Discovery of the Zero-Day Vulnerability
The suspicious activity came to light in September, when Amazon’s team discovered an exposed instance of Cairn, an open-source AI penetration testing framework. Lacking authentication measures, the instance allowed unrestricted visibility into the operator’s commands and the agents’ responses.
The subsequent investigation revealed a two-day attack campaign focusing on IP addresses associated with East Asian universities outside of mainland China. The operator tasked the agents with identifying software vulnerabilities and gaining shell access to the systems.
Exploitation Tactics and Techniques
Upon scrutinizing one target’s software and its API, the AI agents identified a path traversal vulnerability in a file upload mechanism. Amazon disclosed that within just 11 minutes, the agents successfully manipulated the application’s API to execute arbitrary file writes on the server.
Further exploiting this flaw, the agents embedded a Python reverse shell that executed with every activation of the Python interpreter, eventually achieving root shell access. This method enabled them to maintain control over the compromised systems.
Response and Attribution
Amazon’s analysis assessed with high confidence that this was a zero-day vulnerability, as there were no existing patches or advisories for the affected open-source project. The operator directed the agents to use a proxy to infiltrate another university’s network, suggesting the zero-day had been utilized to breach this secondary target as well.
The developers behind MONAI Label issued an advisory, categorizing the bug as critical. It affected versions 0.8.5 and prior, and authentication was typically disabled by default, allowing network-wide exploitation. A patch has been prepared and will be included in the upcoming software release.
Amazon attributes the attack to an unidentified Chinese-linked threat actor, drawing conclusions from the language in the commands, time zone settings, and the strategic significance of the targeted universities to China.
As cyber threats continue to evolve, this incident underscores the need for rigorous cybersecurity measures and prompt response strategies to mitigate the risks associated with emerging vulnerabilities.
