TL;DR
Amazon Threat Intelligence revealed that Chinese-nexus hackers used Cairn AI agents to find and exploit a MONAI Label zero-day in the wild. Tracked as CVE-2026-133968 (CVSS 9.8), this path bug gave root access on school servers in September 2026. Admins must install the Project MONAI fix right away.
- CVE: CVE-2026-133968
- CVSS: 9.8 (Critical · CVSSv3)
- Product: Project-MONIA MONAI Label
- Affected: ≤ 0.8.5
- Impact: CWE-24
- Status: Exploited in the wild
- Action: See vendor advisory
Why It Matters
MONAI Label is a widely used open-source medical scan tool with 912 stars and 281 forks on GitHub. Hospitals, labs, and schools run it on fast GPU hosts to label health scans. Because these hosts store private research data, a root breach gives threat actors deep network access.
More importantly, this event marks a major shift in real-world hacking. In the Amazon Threat Intelligence report, analysts state: “During the campaign, the agents found a zero-day in an open-source application and used it to gain full control of a server. These findings show that agentic offensive operations are already happening, and defenders should prepare for the speed and persistence these tools enable.”
NVD records confirm that hackers actively used this MONAI Label zero-day in the wild starting in September 2026. However, no standalone public exploit script has been shared outside the incident report.
How the Attack Works
AI Agents Run With Cairn
On September 10, 2026, Amazon Threat Intelligence found an open server running Cairn with no login check. Because the operator left it exposed, analysts read full logs from a 48-hour raid on East Asian schools. The actor gave the AI agents target IPs and one simple goal: “obtain shell access.”
First, the agents showed high grit against locked-down hosts. Over 16 hours, they ran 148 steps against an open Redis host and mapped a Docker network via a Qdrant SSRF bug. Next, they spent six hours scraping 80 real staff names from public pages. They then tried over 30,000 logins against a GPU host, a VPN, and a campus site. Although both raids failed, the agents kept adapting with no human help.
Path Bug to Root Shell
Against a host running MONAI Label, the AI agents won fast. Specifically, CVE-2026-133968 is a path traversal bug in the image field of the PUT /datastore/image endpoint. Amazon notes: “In about 11 minutes, the agents went from mapping the application’s API to arbitrarily writing files on the server.”
In a later run, the agents used this MONAI Label zero-day to write a Python reverse shell into sitecustomize.py. Python loads that file each time it starts. Then, they hit a training API path to spawn a new Python task, giving them a root shell. Afterward, the human actor stepped in with a proxy tunnel into a second school network running the same app.
Affected Versions
Per NVD and GitHub Security Advisory, this MONAI Label zero-day hits all MONAI Label builds through 0.8.5.
Patch and Mitigation Steps
Amazon alerted Project MONAI leads, who merged a fix in pull request 1915 and posted GHSA-qc22-rmgc-vg94. Admins should update MONAI Label right away to a fixed build. Also, block direct web access to MONAI Label API ports, run the app as a non-root user, and check Python startup files like sitecustomize.py for rogue code.
Support Our Threat Intelligence
Find our vulnerability reports and weekly recaps helpful? Support our work today and unlock a 100% ad-free reading experience!