“Within 11 minutes, they went from mapping the application's API to exploiting the vulnerability.” That timeline, reported by Amazon Threat Intelligence, captures how an autonomous AI agent platform called Cairn was used to find and weaponize a zero‑day in a medical imaging application and seize full control of a server.
Cairn autonomous AI agents: coordination and a single instruction
Cairn is a legitimate penetration‑testing platform designed to coordinate multiple autonomous AI agents for research and defensive purposes. In the campaign Amazon uncovered, operators issued a succinct order—“obtain shell access”—and left Cairn's agents to probe, discover, and exploit weaknesses on their own. The attackers, however, left their Cairn instance publicly accessible without authentication, exposing the activity and allowing Amazon Threat Intelligence to observe the operation in the clear.
The MONAI Label path‑traversal zero‑day and rapid escalation
Amazon reported the agents discovered a path‑traversal zero‑day in MONAI Label, an open‑source medical imaging application used to help researchers label medical images. The flaw allowed the agents to write files to arbitrary locations on the target server. According to Amazon's timeline, the agents moved from mapping MONAI Label’s API to exploiting the vulnerability in roughly 11 minutes.
In a subsequent session, the agents wrote a Python reverse shell into the sitecustomize.py file—a module that automatically loads when Python starts—and then triggered an API endpoint to gain root access to the server. After discovery, Amazon reported the flaw to MONAI Label maintainers; a security update addressing the issue was released on November 18.

This site is the portfolio.
OSINTSights runs on Cloudflare Workers, D1, R2, and Vectorize, with an AI pipeline on Hetzner ARM. Nubivance designed, built, and operates it. We do the same for clients.
See what we buildObserved attacker behavior, targets, and attribution
Amazon says the campaign was launched between September 8 and 10, 2026, and primarily targeted East Asian educational institutions located outside China. The researchers attributed the campaign to a Chinese‑nexus threat actor with moderate confidence, citing the language of retrieved prompts, activity timestamps, and the choice of targets.
Amazon shared two IP addresses it observed as part of the campaign: 142[.]4.98.145 and 142[.]4.107.34. The report also notes that, after finding the MONAI Label zero‑day on one network, the operator supplied the agents with a proxy into a second university network running the same vulnerable software and directed them to compromise additional machines.
Successes, failures, and brute‑force persistence
The campaign was not uniformly successful. In one case, agents conducted a 16‑hour, 148‑action effort to take control of a Redis server and failed. In another, when targeting a university GPU server, the agents attempted more than 30,000 credential combinations over six hours but did not achieve access. These episodes show both the persistence of autonomous agents and the limits they sometimes encounter.
Other documented Cairn abuse and the dual‑use problem
Cairn’s capabilities are clearly dual‑use. Gambit Security documented cases of Cairn abuse tied to a large‑scale credit‑card stealing operation in September, and Google reported that Cairn had been deployed in an attack against an aerospace organization in the United States. Those incidents, together with Amazon’s finding, illustrate that a tool created “for research and defense strengthening” can be repurposed by attackers to automate discovery and exploitation at scale.
What this means for East Asian universities, MONAI Label maintainers, and security teams
- East Asian universities targeted outside China: Institutions running MONAI Label or similar tooling should assume automated agents will probe exposed services and prioritize applying the November 18 security update and any other patches to vulnerable endpoints.
- MONAI Label maintainers and open‑source maintainers more broadly: The maintainers responded by issuing a security update; their role in rapidly pushing fixes and communicating patches to dependent organizations will materially affect attackers’ timelines.
- Security teams and defenders: Amazon warned that AI agents can find and exploit vulnerabilities faster than security teams can react, and urged organizations to “rely more on automated defenses rather than manual investigations.” Teams may need to integrate automated detection and mitigation to keep pace with autonomous offensive tooling. The two IP addresses Amazon published (142[.]4.98.145 and 142[.]4.107.34) are concrete indicators defenders can use for hunting and network monitoring.
The campaign Amazon observed is notable not only for the technical detail—a path‑traversal write to sitecustomize.py, an 11‑minute escalation, and subsequent root takeover—but for the operational lesson embedded in the attackers’ mistake: leaving the Cairn instance accessible allowed investigators to see both what worked and what failed. The disclosure and the November 18 MONAI Label update are immediate corrective steps; the broader question left by these facts is whether defenders will match attackers’ use of automation with equally rapid, machine‑scale defenses.




