Amazon Details Cairn Agents’ Exploitation of a MONAI Zero-Day

Summary
Amazon researchers say Cairn-driven agents exploited a path-traversal zero-day in the MONAI medical-imaging application, gaining a root reverse shell; activity also targeted a university network. MONAI has patched the flaw and requested a CVE.
Key points
- Amazon said it observed Cairn agents targeting the MONAI medical-imaging application and universities in East Asia.
- The agents reportedly found an upload endpoint vulnerable to path traversal, enabling arbitrary file writes and remote code execution.
- In a later session, the agents wrote a Python reverse-shell payload to sitecustomize.py and triggered a training API endpoint; the resulting process ran as root.
- Amazon said the agents compromised MONAI in 11 minutes and used a tunnel into a university network to reach other machines.
- Amazon found a publicly accessible Cairn instance on September 10 and examined its configurations, prompts, and AI output.
- Amazon attributed the activity to a China-aligned threat actor with moderate confidence, citing Simplified Chinese prompts and China Standard Time timestamps.
- MONAI has issued a patch and requested a CVE identifier; Amazon said other attempts against East Asian universities were unsuccessful.
Article Details
- Attack Vectors
- The attacker instructed Cairn-coordinated AI agents to obtain shell access to the Monai medical imaging application.
- The agents mapped the application API and exploited a path-traversal vulnerability in an upload endpoint's filename parameter to write arbitrary files and execute code remotely.
- The agents wrote a Python reverse-shell payload to sitecustomize.py, then triggered a training API endpoint to start a Python process that imported the payload; the reverse shell ran as root.
- The attacker used an established tunnel through a university network to direct agents toward other machines on that network.
- Defensive Notes
- Monai issued a patch for the zero-day and requested a CVE identifier.
- CJ Moses recommended prioritizing high-confidence detections that trigger automated response, with humans overseeing outcomes.
MITRE ATT&CK
T1059.006 · PythonA Python reverse-shell payload was executed when the agents triggered the training API endpoint to spawn a Python process.T1090.001 · Internal ProxyThe attacker used an established tunnel through a university network to reach other machines on that network.T1190 · Exploit Public-Facing ApplicationAgents exploited a path-traversal flaw in Monai's upload endpoint to write files and execute code on the server.T1546.018 · Python Startup HooksThe agents placed a payload in sitecustomize.py, which Python imports at interpreter startup, and triggered a new Python process to load it.
People
Products
Tools
ARTEXin recent months. Google Mandiant said last month it observed multiple threat actors using Cairn, Artex, PentAGI and Hexstrike to conduct significant parts of the attack life cycle, including lateral movement,CairnCairn is one of many open-source frameworks that coordinates artificial intelligence agents in reaching a goal. In a campaign Amazon detected in September, the attacker told agents harnessed with Cairn to "obtain shellHexstrikeGoogle Mandiant said last month it observed multiple threat actors using Cairn, Artex, PentAGI and Hexstrike to conduct significant parts of the attack life cycle, including lateral movement, database exploitationPentAGIin recent months. Google Mandiant said last month it observed multiple threat actors using Cairn, Artex, PentAGI and Hexstrike to conduct significant parts of the attack life cycle, including lateral movement, database
