Amazon Threat Intelligence says a Chinese-nexus actor used autonomous agents in the Cairn penetration-testing framework to target university networks in East Asia and discover a path-traversal flaw in MONAI Label, an open-source medical-imaging application.
The vulnerability, now tracked as CVE-2026-133968, was used to write files outside the application’s datastore and obtain a root-level shell on a university server, according to Amazon’s November 20 report.
Amazon said it found a publicly accessible Cairn instance on September 10, exposing the operator’s prompts and the agents’ activity. Its records showed a campaign targeting university IP addresses in East Asia outside mainland China over about 48 hours beginning September 8. The operator gave the agents a simple objective — obtain shell access — and let them plan and adjust their activity. On a server running MONAI Label, the agents mapped the application’s API and found an upload endpoint with a caller-controlled image parameter. They reached arbitrary file writes about 11 minutes after mapping the API.
In a later session, the agents used the same flaw to place a Python reverse-shell payload in sitecustomize.py, which Python imports when a new process starts. They then triggered a training API endpoint, causing a new process to load the payload and establish a shell running as root. Amazon said the operator later provided a proxy into a second university network where the same vulnerable software was running. The company inferred that the actor had used the flaw to compromise that server and establish the tunnel, but did not name either university.
The project’s GitHub security advisory GHSA-qc22-rmgc-vg94 lists MONAI Label versions 0.8.5 and earlier as affected. The PUT /datastore/image and PUT /datastore/label endpoints used caller-controlled image IDs and label tags to construct file paths without checking that the destination stayed inside the datastore. Authentication is disabled by default; when it is enabled, the advisory says a user with the ANNOTATOR role can reach the vulnerable functions. The CVE record assigns the flaw a CVSS 3.1 score of 9.8 and says it was exploited in the wild starting in September 2026.
The maintainers merged a path-confinement fix in pull request 1915 on October 19. The advisory says the fix will be included in the next MONAI Label release but does not identify a patched package version. Amazon assessed the zero-day finding with high confidence and its attribution to a Chinese-nexus actor with moderate confidence, citing Simplified Chinese prompts, China Standard Time timestamps and the choice of targets. The company also described separate attempts against a Redis server and a university GPU server that failed to gain access.



