Skip to main content

Key Takeaways

  • Amazon Threat Intelligence reported on November 20, 2026, that AI agents run through the open-source Cairn framework found a previously unknown flaw in MONAI Label and used it to gain full control of a university server in East Asia.
  • Amazon said the Cairn agents went from mapping the MONAI Label application to writing files on its server in about 11 minutes.
  • Amazon assessed with moderate confidence that a Chinese-nexus threat actor ran the September 2026 campaign against universities in East Asia outside mainland China.
  • The MONAI Label security advisory GHSA-qc22-rmgc-vg94, published November 18, 2026, rates the flaw critical for versions 0.8.5 and earlier and says the fix will ship in the next MONAI Label release.

What happened

Amazon Threat Intelligence published findings on November 20, 2026, about a threat actor that used autonomous AI agents to break into university systems with little human involvement. The report was written by CJ Moses, the chief information security officer (CISO) of Amazon Integrated Security.

The actor used Cairn, an open-source AI penetration testing framework. Given a starting point and a goal in plain language, Cairn coordinates several AI agents to reach that goal, Amazon said. The actor gave the agents a list of targets and variations of one goal: "obtain shell access," which The Record described as enough access to a system for various actions to be taken.

During the campaign the agents found a zero-day, a flaw unknown to the software's developers. The flaw is in MONAI Label, an open-source tool for AI-assisted annotation of medical images, SecurityWeek and The Register reported. Amazon said the agents used it to gain full control of a server.

How Amazon found the campaign

On September 10, 2026, Amazon found a Cairn server that ran without authentication, so its history, including the operator's prompts and the agents' output, was visible to anyone on the internet. The records showed a campaign against internet addresses of universities in East Asia outside mainland China over about 48 hours from September 8, Amazon said.

Most of the agents' attempts failed. One operation took about 16 hours and 148 actions against a database server at one university and did not succeed. Another spent more than 6 hours testing more than 30,000 username and password combinations against a university computing server and never found a working login, according to Amazon.

The attempt against MONAI Label succeeded. "In about 11 minutes, the agents went from mapping the application's API to arbitrarily writing files on the server," Amazon said. The agents later used the same flaw to run code on the server with root, the highest level of access, Amazon said.

Amazon said it assessed with high confidence that the flaw was a zero-day, because no fix, issue, advisory, or earlier mention of it existed. The actor then gave the agents a tunnel into a second university network. The tunnel ran through a server with the same software, and Amazon inferred that the actor had used the zero-day to break into that server.

Who is behind it

Amazon assessed with moderate confidence that the actor is "Chinese-nexus," based on prompts written in simplified Chinese, scripts set to China Standard Time, and targets "likely of strategic interest to China." AWS declined to name the universities or the country they are in, The Record reported.

The affected universities were not AWS customers, and AWS worked with a government partner to notify them, Amazon said. Amazon published indicators of compromise in its report.

Who is affected and what to do

The MONAI Label developers published advisory GHSA-qc22-rmgc-vg94 on November 18, 2026. It rates the flaw critical and lists versions 0.8.5 and earlier as affected. The flaw is a path traversal, a bug that lets a request write files outside the folder the software intends, according to the advisory.

Authentication is disabled by default in MONAI Label, so any network client that can reach a server can exploit the flaw, the advisory says. The developers have fixed the code and will include the fix in the next MONAI Label release, the advisory says. It lists no patched version yet.

Until an update is available, the advisory recommends turning on authentication and not giving the annotator role to untrusted users. It also advises keeping MONAI Label servers off untrusted networks and running them as an unprivileged user that can only write to the datastore folder.

Amazon advised organizations to rely on high-confidence detections that trigger automated responses, with people overseeing the outcomes. An agent that finds and exploits a zero-day in minutes "completes its activity faster than many security operations centers can manually triage an alert," Amazon said.

Background

Amazon cited earlier research by Gambit Security describing a separate, financially motivated actor that used Cairn against online retailers. "Together, these cases show that this autonomous AI agent framework is being adopted in both espionage and criminal operations," Amazon said.

Google researchers also reported on November 17 that they saw Cairn used in intrusions at companies in the US aerospace sector, according to The Record. The FBI issued an alert on November 13 about Artex, another AI framework that attackers use to automate intrusions, The Record reported.