Summary
Amazon Threat Intelligence reports that a threat actor used the open-source AI penetration-testing framework Cairn to attack East Asian universities in September 2026. In one case the AI agents discovered a previously unknown path traversal flaw in MONAI Label, an open-source medical imaging application, and used it to take over a server. Amazon attributes the activity with moderate confidence to a Chinese-nexus actor.
What happened
Cairn coordinates multiple AI agents. The operator told them to "obtain shell access" and let them work out the rest.
- Dates: the attacks ran September 8 to 10, 2026.
- Zero-day: the agents found a path traversal bug in MONAI Label that allows file writes to arbitrary locations on the server.
- Speed: they went from mapping the API to exploiting the flaw in about 11 minutes.
- Root access: the agents wrote a Python reverse shell into
sitecustomize.py, then called an API endpoint to trigger it and gain root. - Lateral movement: the operator then gave the agents a proxy into a second university network running the same vulnerable software.
Not everything worked. A 16-hour, 148-action attempt to take over a Redis server failed, and more than 30,000 credential combinations tried over six hours against a university GPU server did not succeed.
Attribution and discovery
Amazon bases the China-nexus assessment on the language of the retrieved prompts, activity timestamps and target choices. Amazon saw the operation because the attackers left their Cairn instance publicly reachable without authentication.
Cairn has been abused before. Gambit Security documented its use in a September card theft operation, and Google reported it used against a US aerospace organization.
Patch status
Amazon reported the flaw to the MONAI Label maintainers, who released a fix on November 18, 2026 (advisory GHSA-qc22-rmgc-vg94).
Indicators
Amazon shared two IP addresses seen September 8 to 10: 142[.]4.98.145 and 142[.]4.107.34.
What to do
- If you run MONAI Label, update to the November 18 release now and keep it off the open internet.
- Search logs for connections from the two IPs above and for unexpected changes to
sitecustomize.pyin the application's Python environment. - Restrict write paths for the service account so a file-write bug cannot reach interpreter startup files.
- Plan for faster exploitation. Amazon warns that AI agents can find and exploit flaws faster than teams can investigate by hand, and urges more automated defenses.