Amazon's threat intelligence team observed autonomous AI agents, operated by a China-linked actor, uncovering a zero-day flaw in MONAI Label (SecurityWeek). The open‑source tool assists AI‑driven annotation of medical images. Researchers first spotted the activity in September after finding an exposed Cairn instance—a penetration‑testing framework—running without authentication.
Because the Cairn instance lacked authentication, researchers could see the operator's prompts and the agents' output. Log analysis showed a roughly 48‑hour campaign aimed at IP addresses belonging to universities in East Asia, outside mainland China. The operator fed target lists to the agents and instructed them to obtain shell access.
During the operation, the agents identified MONAI Label on a target host, mapped its API, and located a path‑traversal bug in a file‑upload function. Amazon reports that the transition from API mapping to arbitrary file write took about 11 minutes.
In a follow‑up session, the agents used the flaw to drop a Python reverse shell into a file that runs automatically whenever the interpreter starts. This step gave the threat actor a root shell on the compromised system.
Amazon's intelligence unit judged the vulnerability to be a zero‑day, citing the lack of prior advisories, patches, or database entries. The actor then directed the agents to tunnel through a proxy and reach additional devices inside a second university's network.
MONAI Label developers issued a security advisory confirming the critical issue affects version 0.8.5 and earlier. They noted that authentication is disabled by default, allowing any network‑adjacent user to exploit the bug. An official patch is slated for the project's next release.
Amazon attributes the campaign to an unidentified Chinese‑nexus actor with moderate confidence. The assessment rests on script time zones, prompt language choices, and the strategic relevance of the targeted academic institutions.
The rapid 11‑minute transition from API mapping to arbitrary file placement suggests that offensive AI frameworks can drastically compress the time window defenders have to respond to zero‑day exposures. Because MONAI Label ships with authentication turned off by default, the incident highlights how permissive default configurations in open‑source AI utilities likely exacerbate the risk of automated discovery. Moving forward, security teams will likely need to implement stricter network segmentation around specialized open‑source tools to mitigate exposure from autonomous agent scanning.
Spotted an error? See our corrections policy or email editorial@lordofclaude.com.





Discussion & Comments (0)
Join the conversation