AEGLIX News
Suspected China-linked hackers deployed autonomous AI agents to discover and exploit a previously unknown path-traversal flaw in MONAI Label, gaining root access to a university server in September, according to Amazon Threat Intelligence. The vulnerability, tracked as CVE-2026-133968, affects MONAI Label versions 0.8.5 and earlier in their default configuration where authentication is disabled, allowing unauthenticated network clients to write files outside the designated storage folder.
The flaw resides in the PUT /datastore/image endpoint where a caller-supplied file identifier is appended to the storage path without validation that the resulting path remains within the approved directory. This directory traversal weakness enables remote code execution via the image parameter. The CVE record published this week assigns a critical severity rating of 9.8 out of 10. Amazon observed the AI agents mapping the application interface and achieving file write capability in approximately 11 minutes over a 48-hour period.
In a separate session, the AI agents leveraged the same vulnerability to plant a file that Python executes automatically at startup, then triggered a training feature that loaded the malicious file and established a reverse connection to the attacker with root privileges. This compromise occurred on one university server, though Amazon notes that whether another MONAI Label installation would yield identical root access depends on the service account configuration and deployment specifics. Amazon also detected a later session where an operator provided the agents a tunnel into a second university network running the same software, believing it was compromised similarly but without confirmation.
The MONAI Label maintainers confirmed the vulnerability and merged a source code fix on October 19 that forces uploads to resolve inside approved folders while rejecting empty filenames or names beginning with a dot. The patch addresses the path-traversal issue by implementing proper path validation at the datastore endpoint. Amazon Threat Intelligence disclosed the exploitation activity and the maintainers' remediation, with the CVE record now publicly available documenting the directory traversal and resultant remote code execution exploited in the wild starting in September 2026.
Report prepared 2026-11-21 · Based on reporting from The Hacker News. Source publication date above; subsequent developments may not be reflected.