Skip to content
See the World Through ScienceA project of ALLATRA

AI Agents Found a Flaw in Medical Imaging Software on Their Own, Amazon Says

AI & Technology

Republish this story

Our work is licensed under Creative Commons BY-NC 4.0. You may republish this piece for free — with credit to ALLATRA Media and a link to the original, unedited beyond length trims, and not for commercial use.

Read the full license

Rows of rack-mounted servers with blue and red status lights in a darkened equipment room.
Rack-mounted servers at a research institution's headquarters. The agents took control of a single university server, the one running the medical imaging software (illustrative)."NOIRLab HQ Server Racks (6V6A0404-CC)" by NOIRLab/NSF/AURA/T. Slovinský, via Wikimedia Commons, CC BY 4.0, cropped

Amazon Threat Intelligence reported Nov. 20, 2026, that an attacker's AI agents found a previously unreported flaw in open-source medical imaging software and used it to take control of a university server. The attacker gave the agents a simple goal in plain language, to get command-line access to the target, and the agents worked out the steps themselves.

In its account of the campaign, Amazon says it could see the whole operation because the attacker had left their own server open to the internet, with no login required. That server ran Cairn, an open-source framework that coordinates AI agents toward a goal stated in plain language.

Amazon says the agents identified the imaging application's framework, mapped its interfaces and tested a file-upload setting, then went from that mapping to writing files anywhere on the server in about 11 minutes. In a later session they used the same weakness to take control of the server as its most privileged user. Two other operations failed outright.

The affected software is MONAI Label, an open-source tool for labeling medical images to train AI models. Its maintainers published a security advisory on Nov. 18, 2026, that rates the flaw Critical and lists versions 0.8.5 and earlier as affected. It lists no patched release and no CVE identifier, and notes that authentication is off by default.

The two failed operations ran far longer — hours rather than minutes. Amazon says the most extensive worked for roughly 16 hours and 148 sequential actions against a university database left open without a password, and got nowhere. Another tested more than 30,000 login combinations against a university research server, using usernames the agents had collected from faculty directories, lab websites and papers. None worked.

Amazon says it assesses with high confidence that the flaw had not been reported before, because no fix, report or advisory for it existed in vulnerability databases. The company found the operator's server in September 2026, during a campaign against university addresses in East Asia.

Amazon notified the project's maintainers. It also points to research by Gambit Security on a separate, financially motivated group using Cairn against online retailers.

Sources

Spot an error?

Spot an error?

Report an error

Spotted a mistake on this page? Tell us what's wrong and our editors will take a look.

What kind of problem?

Only if you'd like us to be able to follow up. We won't use it for anything else.

We correct mistakes openly. Select any text to flag it. Fixes are logged under our Corrections Policy.

Report an error

Reporting on

AI Agents Found a Flaw in Medical Imaging Software on Their Own, Amazon Says

What kind of problem?

Only if you'd like us to be able to follow up. We won't use it for anything else.

We read every report. Corrections are logged publicly.