Autonomous AI Agents Exploit MONAI Label Vulnerability in University Cyberattack

Share

- Advertisement -

Artificial intelligence is changing the cybersecurity landscape, and not always for the better. While security teams increasingly use AI to identify vulnerabilities and defend networks, cybercriminals are beginning to use the same technology to find weaknesses and break into systems with far less human involvement.

A recent cyberattack involving an open source medical imaging platform has highlighted just how serious this development could become. According to Amazon Threat Intelligence, suspected China linked attackers used autonomous AI agents to discover a previously unknown vulnerability in MONAI Label, a medical imaging server used by research teams.

The attackers reportedly exploited the flaw to gain access to a university server and eventually obtain root privileges. What makes the incident particularly concerning is how independently the AI agents operated. Rather than relying on detailed instructions at every stage, they explored the application, identified a weakness and worked towards gaining access with limited human guidance.

The incident offers a worrying glimpse into how AI powered hacking could evolve, particularly when vulnerable software is exposed to the internet.

AI Agents Discovered a Critical Vulnerability in MONAI Label

The vulnerability affects MONAI Label, an open source application designed to help researchers annotate medical images and support the development of medical imaging AI models.

According to Amazon’s investigation, the security weakness was present in MONAI Label version 0.8.5 and earlier. The problem involved a path traversal vulnerability in the application’s file handling functionality.

- Advertisement -

In simple terms, the software failed to properly check where uploaded files were being saved. An attacker could manipulate a file identifier to make the server write files outside its intended storage directory.

This is particularly dangerous because the default MONAI Label configuration does not enable authentication. As a result, anyone who could reach an exposed server over the network could potentially attempt to exploit the vulnerability without having a legitimate account.

The flaw did not automatically give attackers complete control over an entire machine. However, the ability to write files to sensitive locations could create opportunities for further exploitation, depending on the server’s configuration and the permissions available to the application.

The vulnerability has been assigned CVE-2026-133968 and received a critical severity score of 9.8 out of 10.

How Autonomous AI Agents Carried Out the Attack

One of the most significant aspects of this incident is the role played by autonomous AI agents.

Amazon reconstructed the operation after discovering an exposed Cairn control panel belonging to the suspected attackers. Cairn is an open source framework designed to run AI agents that perform automated penetration testing.

- Advertisement -

The exposed control panel provided investigators with valuable information, including the agents’ prompts, configuration settings and previous activity.

According to Amazon, the operation continued for approximately 48 hours in September. During that period, the AI agents explored the target application’s interface and identified a way to write files to the server.

The agents reportedly reached this stage in approximately 11 minutes.

However, this does not mean they gained root access within those 11 minutes. That distinction is important because the initial file writing capability and the later privilege escalation were separate stages of the attack.

The operator provided the overall objective, selected the targets and instructed the agents to obtain shell access. The AI agents then chose their own methods and carried out much of the technical exploration independently.

In a separate session, they exploited the same vulnerability to place a file in a Python startup location. They subsequently triggered a training feature that launched a new Python process.

- Advertisement -

When that process started, it automatically loaded the planted file and established a connection back to the attackers with root privileges.

This gave the attackers root level access to one university server. However, the result depended on the way the software was deployed and the permissions available to its service account. It should not be assumed that every MONAI Label installation would be vulnerable to the same level of compromise.

China Linked Activity Raises Further Security Concerns

Amazon attributed the campaign to a suspected China linked threat actor with moderate confidence.

Investigators based this assessment on several indicators, including Simplified Chinese language prompts, China Standard Time settings and the selection of targets that could be of interest to Chinese operators.

However, Amazon did not associate the operation with a specific hacking group or establish that a government was directly involved.

The investigation also uncovered activity involving a second university network. During a later session, the operator provided the AI agents with a tunnel into another network running the same software.

Amazon believes the attackers may have compromised that environment using a similar approach, although the second intrusion has not been independently confirmed.

It is also important to understand what the investigation did not establish. There is no confirmed evidence that patient scans or medical records were stolen, or that the attackers accessed clinical systems beyond the university environments described.

Nevertheless, the incident demonstrates why organisations running specialist research software need to treat exposed services as potential entry points for wider attacks.

No Patched Release Available Yet: What Organisations Should Do

The vulnerability presents an additional challenge because the developers have already merged a security fix into the MONAI Label source code, but a patched software release is not yet available.

The maintainers merged the correction on October 19. The change ensures uploaded files remain within approved storage directories and rejects invalid file names.

However, the project’s security advisory still does not list a patched version. The latest published package remains version 0.8.5, meaning organisations cannot simply install a newer official release and assume the problem has been resolved.

For organisations currently using MONAI Label, reducing exposure should be the immediate priority.

First, administrators should restrict access to MONAI Label servers through firewall and network routing rules. Leaving an exposed service accessible to untrusted users creates an unnecessary security risk.

Second, authentication should be enabled wherever possible. Since the default configuration does not require users to log in, administrators should avoid relying on network isolation alone.

Third, the application should never run with unnecessary administrative privileges. Running MONAI Label under a dedicated, restricted service account can significantly reduce the damage an attacker might cause after exploiting a vulnerability.

Organisations should also review the directories the application can access and modify. Limiting write permissions makes it harder for an attacker to turn a file upload vulnerability into code execution or broader system compromise.

Finally, administrators should investigate systems that were previously exposed to the internet. Unexpected Python startup files, unfamiliar training jobs and unusual outbound network connections deserve particular attention.

A source code fix is useful, but organisations should verify that their deployment actually includes the correction before considering the vulnerability addressed.

The wider lesson is difficult to ignore. AI agents are becoming capable of performing increasingly complex cybersecurity tasks, including identifying vulnerabilities and chaining together multiple steps to gain access to systems.

For defenders, this creates a new challenge. Security teams are no longer dealing exclusively with human attackers who manually explore applications and develop exploits. They may also face automated agents capable of conducting reconnaissance and testing weaknesses at considerable speed.

The MONAI Label incident does not suggest that AI can independently compromise every vulnerable system. However, it demonstrates that autonomous agents can meaningfully accelerate real world intrusion attempts.

Follow TechBSB For More Updates

- Advertisement -
Emily Parker
Emily Parker
Emily Parker is a seasoned tech consultant with a proven track record of delivering innovative solutions to clients across various industries. With a deep understanding of emerging technologies and their practical applications, Emily excels in guiding businesses through digital transformation initiatives. Her expertise lies in leveraging data analytics, cloud computing, and cybersecurity to optimize processes, drive efficiency, and enhance overall business performance. Known for her strategic vision and collaborative approach, Emily works closely with stakeholders to identify opportunities and implement tailored solutions that meet the unique needs of each organization. As a trusted advisor, she is committed to staying ahead of industry trends and empowering clients to embrace technological advancements for sustainable growth.

Read More

Trending Now