Amazon Says AI Agents Used by China-Linked Actor Exploited MONAI Label Zero-Day

Summary
Amazon says AI agents used in a roughly 48-hour campaign exploited a MONAI Label path traversal flaw to gain root access on university systems. Developers rate the bug critical and say a fix is planned for the next release.
Key points
- Amazon says its threat intelligence team found records of the activity in September on an exposed, unauthenticated Cairn penetration-testing framework instance.
- The operator reportedly directed agents to target university IP addresses in East Asia outside mainland China and obtain shell access.
- The agents reportedly found a path traversal flaw in MONAI Label’s file-upload function and wrote files to a server within about 11 minutes.
- In a later session, the agents reportedly planted a Python reverse shell, ultimately gaining root access.
- Amazon inferred that the actor used the same flaw to compromise a second university server and establish a proxy tunnel; this was not described as independently confirmed.
- MONAI Label developers rate the flaw critical, say it affects versions 0.8.5 and earlier, and report that a fix will be included in the next release. Authentication is off by default.
- Amazon assesses with high confidence that the flaw was a zero-day and with moderate confidence attributes the activity to an unidentified China-linked actor.
Article Details
- Event Type
- AI-agent-assisted zero-day exploitation of MONAI Label, attributed by Amazon with moderate confidence to an unidentified China-linked threat actor.
- Impact
- Amazon reports that the agents exploited a path traversal vulnerability in MONAI Label, wrote files to a university server, and achieved root shell access. Amazon infers that a second university server was compromised and used to establish a tunnel. The developers rate the vulnerability critical; it affects MONAI Label versions 0.8.5 and earlier, is exploitable without authentication when the server is network-reachable, and has a fix planned for the next release.
