JupiteX Get the app
Science & Technology21 Nov 2026 · about 6 min

Hackers use autonomous AI tool Cairn to find zero-day, take over server

The brief

The agents found a path-traversal zero-day in MONAI Label, an open-source medical imaging application that helps researchers label images with AI. A zero-day is a flaw not yet addressed by a security update. This discovery matters because the flaw allowed file writes beyond the application’s intended locations. Amazon reported that the agents went from mapping MONAI Label’s API to exploiting the flaw in 11 minutes. In a later session, they used it to write a Python reverse shell into sitecustomize.py, a file that automatically loads when Python starts. They then triggered an API endpoint to establish root access to the server. The activity was part of a campaign targeting mainly East Asian educational institutions outside China. Amazon reported the flaw to MONAI Label’s maintainers, who released a security update on November 18. The case shows how quickly an AI agent can move from examining an application to using a vulnerability, making timely fixes and defenses important.

01

What flaw did Cairn’s agents find in MONAI Label, and what did it let them do?

The agents found a path-traversal zero-day in MONAI Label, an open-source medical imaging application that helps researchers label images with AI. A zero-day is a flaw not yet addressed by a security update. This discovery matters because the flaw allowed file writes beyond the application’s intended locations.

Amazon reported that the agents went from mapping MONAI Label’s API to exploiting the flaw in 11 minutes. In a later session, they used it to write a Python reverse shell into sitecustomize.py, a file that automatically loads when Python starts. They then triggered an API endpoint to establish root access to the server.

The activity was part of a campaign targeting mainly East Asian educational institutions outside China. Amazon reported the flaw to MONAI Label’s maintainers, who released a security update on November 18. The case shows how quickly an AI agent can move from examining an application to using a vulnerability, making timely fixes and defenses important.

02

What is a path-traversal vulnerability, and how can it let software write files where it should not?

A path-traversal vulnerability happens when software handles a file path in a way that lets a user reach outside the intended folder. In general, attackers may manipulate path details, such as using sequences that move up through directories. The key problem is that the application does not properly limit where a requested file operation can go.

Amazon said a path-traversal zero-day in MONAI Label let Cairn’s agents write files to arbitrary locations on the server. They used the flaw to place a Python reverse shell in sitecustomize.py, which loads automatically when Python starts. They then triggered an API endpoint, establishing root access in that session.

The report does not describe the exact path string or the application’s vulnerable code, so those details cannot be inferred. Amazon reported the issue to MONAI Label maintainers, who released a security update on November 18. Restricting file access to approved locations and applying fixes can reduce exposure to this kind of flaw.

03

How quickly did the agents go from mapping MONAI Label’s API to exploiting the flaw?

The agents took 11 minutes to move from mapping MONAI Label’s API to exploiting a path-traversal zero-day. That speed matters because it shows how quickly an automated operation can progress from learning how an application behaves to using a flaw against it.

MONAI Label is an open-source medical imaging application that uses AI to help researchers label medical images. The vulnerability let the agents write files to arbitrary locations on the server. In a subsequent session, they wrote a Python reverse shell to sitecustomize.py, which automatically loads when Python starts, and triggered an API endpoint to establish root access.

The 11-minute figure covers the mapping-to-exploitation step, not the entire campaign or every action that followed. Amazon later reported the flaw to MONAI Label maintainers, who released a security update on November 18. Amazon warned that AI agents may find and exploit vulnerabilities faster than security teams can react.

04

What is Cairn, and what made its AI agents autonomous in this operation?

Cairn is a legitimate pentesting tool created for research and defense strengthening. It can coordinate multiple AI agents in cyber-operations. Its intended purpose is security testing, but the article reports that attackers used the platform in a campaign against vulnerable servers.

In this operation, the suspected attackers instructed the agents to “obtain shell access.” Rather than being given every step, the agents independently explored targets and discovered a path-traversal zero-day in MONAI Label. They mapped its API, exploited the flaw, wrote a Python reverse shell, and triggered an endpoint to establish root access.

The article describes autonomy as the agents’ ability to carry out the task and discover and exploit the flaw independently. Their work was not uniformly successful: a 16-hour, 148-action attempt against a Redis server failed, as did over 30,000 credential attempts against a university GPU server. The campaign illustrates that autonomous tools can make progress without succeeding at every target.

05

How did writing a reverse shell to sitecustomize.py help the attackers gain root access?

A reverse shell is a tool that can give an attacker a way to control a compromised system. The important detail in this case is where the agents placed it: sitecustomize.py. According to Amazon, that file automatically loads when Python starts, giving the written code a chance to run during Python startup.

In a subsequent session, the agents used the MONAI Label path-traversal flaw to write the Python reverse shell to that file. They then triggered an API endpoint, which the report says established root access to the server. This followed an earlier session in which the agents mapped the application’s API and exploited the flaw.

The source does not give further technical details about the endpoint or the exact sequence by which root privileges were obtained, so those specifics are unknown. It does establish that arbitrary file writing enabled the placement of the shell and that the endpoint trigger led to root access. The incident shows why unrestricted file writes can create serious security consequences.

06

What can an attacker do after gaining root access to a server?

Root access is the highest level of administrative control on a server. In general, an attacker with that level of access may be able to change system settings, run programs, manage accounts, or reach data available on the machine. The article does not list the attackers’ actions after gaining root, so those possibilities should not be mistaken for reported activity.

In the MONAI Label case, Cairn’s agents used a path-traversal flaw to write a Python reverse shell to sitecustomize.py. That file automatically loads when Python starts. The agents then triggered an API endpoint, and Amazon said this established root access to the server. The article describes this as taking complete control of a server.

The campaign targeted mainly East Asian educational institutions outside China, and the operator later provided a proxy into a second university network running the same vulnerable software. Amazon reported the flaw, and maintainers released an update on November 18. The episode shows why a server compromise can matter beyond the initial vulnerable application.

07

Why are zero-day flaws difficult to defend against, and how can security updates and automated defenses reduce the risk?

A zero-day is a vulnerability that attackers can exploit before a security update addresses it. That creates a difficult window for defenders: the flaw may be used before a fix is ready or applied. In the MONAI Label case, the agents found a path-traversal flaw and used it to write files to arbitrary server locations.

Amazon reported the flaw to MONAI Label maintainers, who released a security update on November 18. Applying available security updates can close a known vulnerability on systems that receive the fix. Amazon also 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.

The article does not specify particular defensive products or steps. It does show both the value and limits of speed: agents reached exploitation in 11 minutes, while defenders later reported and patched the flaw. Automated defenses may help organizations respond faster, but the source does not claim they guarantee protection.

This brief was written by AI from the original reporting and checked by other models. Names, figures and quotes come from the source; read it for full context.

Read more in the JupiteX app

Pulse is free. New stories every 4 hours, each one broken into the questions that explain it.

Or read more news on the web