Autonomous artificial intelligence agents directed by a suspected China-linked threat actor identified and exploited a zero-day vulnerability in open-source medical software, according to research published by Amazon's threat intelligence team.

The campaign targeted servers running MONAI Label, an open-source application used for AI-assisted annotation of medical imaging data. During the intrusion, automated agents went from initial API reconnaissance to arbitrary file execution in a matter of minutes, ultimately escalating their foothold to gain root shell access on academic networks.

Exposed Pentesting Framework Reveals 48-Hour Campaign

Amazon's researchers uncovered the activity in September after locating an exposed installation of Cairn, an open-source framework designed for AI-driven penetration testing. The server lacked authentication mechanisms, enabling investigators to view the threat actor's exact prompts and the corresponding outputs produced by the automated agents.

According to reporting from SecurityWeek, the operator orchestrated a roughly 48-hour operation against IP addresses assigned to higher education institutions in East Asia outside mainland China. The human controller supplied the autonomous agents with target lists and specific instructions to obtain shell access on the remote machines.

Amazon attributed the intrusions to an unidentified Chinese-nexus threat group with moderate confidence. The assessment relies on the language observed in operator prompts, script time zone indicators, and the geopolitical profile of the targeted academic institutions.

Rapid Discovery and Root Privilege Escalation

Once deployed against an initial victim, the autonomous agents identified the underlying MONAI Label installation, mapped its exposed application programming interfaces, and identified a path traversal flaw residing within a file upload routine.

The speed of the automated reconnaissance and exploitation was notable. Amazon observed that the agents progressed from mapping the software's API to executing arbitrary file writes onto the underlying server in approximately 11 minutes. Amazon assessed the flaw as a genuine zero-day vulnerability with high confidence, noting the absence of any existing patches, public advisories, or entries in vulnerability databases at the time of the event.

In subsequent sessions, the AI tools leveraged the arbitrary file write capability to insert a Python reverse shell into a file that executes automatically each time the host system's Python interpreter launches. This persistence mechanism ultimately yielded root access. The operator later provided the agents with proxy configurations tunneling into a second university system running the same software, instructing the agents to compromise the host and pivot deeper into internal network assets.

Critical Vulnerability in Default Deployments

The project maintainers behind MONAI Label issued a security advisory following notification of the flaw. The developers classified the bug as critical, confirming that it impacts MONAI Label versions 0.8.5 and older.

A primary contributing factor to the vulnerability's severity is that authentication is disabled by default in affected versions. Consequently, any attacker capable of establishing network connectivity to the application's interface can interact with the upload functions and trigger the path traversal vulnerability without providing credentials.

The developers have engineered a fix to remediate the path traversal bug and stated that the patch will be incorporated into the next official software release.

What You Should Do

Organizations hosting MONAI Label—particularly research hospitals, academic institutions, and biomedical laboratories—must take immediate defensive measures to secure their installations:

  • Restrict Network Exposure: Ensure MONAI Label instances are not accessible from the public internet. Place instances behind internal virtual private networks (VPNs) or restrict access using strict firewall access control lists (ACLs).
  • Implement Access Controls: Enable and configure mandatory authentication mechanisms on all deployments, overriding the default unauthenticated state.
  • Monitor Python Execution Paths: Audit system directories and Python startup files for unauthorized modifications or newly created reverse shells, especially in environments running version 0.8.5 or earlier.
  • Inspect Inbound Network Traffic: Review web server and API logs for abnormal file upload requests or directory traversal sequences directed at MONAI Label endpoints.
  • Prepare to Update: Track the MONAI Label project repository and apply the forthcoming patched release immediately once made available by the maintainers.