Hackers believed to be linked to China have leveraged autonomous AI agents to exploit a zero-day vulnerability in MONAI Label, an open-source medical imaging platform. This flaw enabled them to gain root access to a university server, as reported by Amazon Threat Intelligence this week.
Details of the Software Vulnerability
The vulnerability pertains to MONAI Label versions 0.8.5 and earlier, wherein attackers could write files outside the designated server storage without requiring user authentication. This issue arises from a path-traversal flaw in the system’s handling of uploaded images and labels, risking unauthorized code execution.
While the developers have addressed this flaw in the source code, a patched version has yet to be released for public installation. The server, often used for labeling medical scans in research, operates without authentication by default, making it susceptible to network-based exploits.
AI Agents and Their Role in the Breach
The intrusion was reconstructed by Amazon after discovering the hacker’s control panel, Cairn, exposed online without a password. Cairn facilitates AI-driven penetration testing, and its logs revealed the agents’ activities, which included mapping and exploiting the software’s interface in just 11 minutes.
These AI agents operated with minimal guidance, autonomously selecting methods to achieve shell access on the targeted systems. Although the initial breach did not immediately result in root access, subsequent sessions allowed the attackers to exploit the same flaw further.
Broader Implications and Security Measures
The attackers’ activities have been tentatively linked to Chinese interests due to language and time settings, although no specific group or government involvement has been confirmed. The compromised server belonged to a university, and similar setups elsewhere could be at risk depending on their configurations.
Security experts advise restricting network access to MONAI Label servers, enabling authentication, and running the servers under restricted accounts. These measures are crucial in minimizing exposure and potential exploitation.
With the flaw now recognized as CVE-2026-133968 and rated critically at 9.8 out of 10, it is vital for organizations using MONAI Label to implement these protective steps while awaiting an official patch.
Conclusion
While Amazon’s findings have not been independently verified, they highlight significant vulnerabilities in medical imaging systems. Organizations must remain vigilant and proactive in securing their systems against such sophisticated cyber threats.
