Saturday, November 21, 2026

News

Cairn AI Agents Found a Zero-Day in 11 Minutes, Then Took Over a University Server

AI AgentsPatryk Raba
Cairn AI Agents Found a Zero-Day in 11 Minutes, Then Took Over a University Server
Fot. ProjectManhattan, Wikimedia Commons (CC BY-SA 3.0)

Amazon Threat Intelligence describes an attack in which AI agents running in the Cairn tool found an unknown flaw in MONAI Label on their own and gained root access. About 11 minutes passed between mapping the API and writing files on the server.

Contents
  1. What the agents did
  2. Context and limits
  3. How the campaign was found
  4. What it means for companies

Autonomous AI agents launched by suspected Chinese operators found a previously unknown vulnerability in a medical image labeling application on their own and took over a university server. According to Amazon Threat Intelligence, about 11 minutes passed between mapping the API and being able to write files on the server. The operators gave the agents only one instruction: obtain shell access.

What the agents did

The attackers used Cairn, a penetration-testing tool that coordinates multiple AI agents. According to BleepingComputer, it is a legitimate tool built for research and strengthening defenses, but it has also been used in attacks before. The operators told the agents only to obtain shell access on the target server.

The target was MONAI Label, an open-source tool in which AI helps researchers label medical images. The flaw is a path traversal, a bug that allows files to be written to arbitrary locations on the server. In a later session, the agents wrote a Python reverse shell to sitecustomize.py, a file that loads automatically when Python starts, then triggered the relevant API endpoint and gained root access.

Context and limits

Amazon assesses with high confidence that this was a zero-day, because it found no fix, issue or entry in vulnerability databases for it.

After the successful break-in, the operator gave the agents a proxy into the network of a second university running the same vulnerable software and told them to attack further machines. In the same campaign, the agents failed against a Redis server, despite trying for 16 hours and 148 sequential actions. On the GPU server and other university systems, more than 30,000 credential combinations tested over more than six hours did not help them either. Not every attempt succeeded.

How the campaign was found

Amazon picked up the trail because the attackers left their Cairn instance publicly accessible without authentication, which exposed how the operation unfolded. The attacks mainly targeted educational institutions in East Asia outside China. Attributing the case to a China-linked actor rests on the language of the recovered prompts, the activity timestamps and the choice of targets, and Amazon rates it with moderate confidence.

The BleepingComputer article gives neither the names of the universities nor a CVE identifier. The MONAI Label maintainers, notified by Amazon, published a security update on November 18 along with a GitHub Security Advisory. About two months passed since the attacks.

What it means for companies

Amazon warns that an AI agent that finds and exploits a flaw in minutes works faster than many security teams can manually triage an alert. It therefore recommends relying more on automated defenses than on manual investigation of incidents.

For Polish universities and research institutions the practical takeaway is simple: anyone running MONAI Label should check whether they have the November 18 fix. The case is another example of Cairn being used in attacks, which earlier reports had already covered.

Share: