Skip to main content
All incident reports
Incident report

Amazon: AI agents found a MONAI Label zero-day and gained root at a university

Autonomous AI agents run by a suspected China-linked operator found and exploited a previously unknown flaw in MONAI Label, an open-source medical imaging server, and obtained root access to one university server, according to Amazon Threat Intelligence. The maintainers have merged a source-code fix but have not yet published a version users can install.

· First published

The short version

  • The finding. Amazon Threat Intelligence says AI agents run through the open-source Cairn framework discovered a path traversal flaw in MONAI Label and used it to write files on a university server. The flaw affects version 0.8.5 and earlier, and authentication is off by default.
  • How it was traced. Amazon found the operator's Cairn control panel exposed on the internet without a password, which revealed the agents' prompts, task settings and run history. The records showed a roughly 48-hour campaign in September against universities in East Asia outside mainland China.
  • Root access. After mapping the application's interface and writing files in about 11 minutes, the agents later planted a Python startup file and triggered a training job that ran it, returning a root shell on one university server. Amazon says a second university network was probably breached the same way but has not confirmed it.
  • The fix. Maintainers confirmed the flaw and merged a source fix on 19 October, but their advisory still lists no patched release; the newest version remains 0.8.5 from 2024. The CVE record, CVE-2026-133968, rates the flaw 9.8 out of 10.
  • Attribution. Amazon attributes the campaign to an unidentified China-linked actor with moderate confidence, based on Simplified Chinese prompts, China Standard Time settings and the choice of targets. It named no group and did not establish government involvement.

Timeline

DateEvent
SeptemberAmazon detects a roughly 48-hour campaign in which AI agents target university IP addresses in East Asia and reach file writing on one server in about 11 minutes.
19 OctoberMONAI Label maintainers confirm the flaw and merge a source-code fix that forces uploads to resolve inside approved folders.
NovemberThe maintainers publish a critical advisory with no patched version listed, and CVE-2026-133968 is published with a severity score of 9.8.

How the intrusion unfolded

Amazon Threat Intelligence found the campaign after coming across an exposed Cairn control panel with no authentication. Cairn is an open-source framework that runs AI agents as automated penetration testers, and the unprotected panel exposed the operator's prompts, task settings and run history. The records showed a roughly 48-hour campaign in September against IP addresses belonging to universities in East Asia outside mainland China. The operator set the goal of shell access and chose the targets; the agents selected the methods themselves.

On one target the agents identified MONAI Label, mapped its API and found a path traversal bug in a file upload function. A path traversal flaw lets a caller write files outside the folder the software intends to use. Amazon said the agents went from mapping the API to writing files in about 11 minutes. In a later session they planted a Python reverse shell in a file the interpreter runs at startup, then triggered a training feature that started a new Python process and returned a root shell.

The operator then gave the agents a proxy tunnelling into a second university network running the same software. Amazon infers that network was compromised the same way but says it has not confirmed the second intrusion. The company's account rests on its own reconstruction and has not been independently confirmed.

The flaw and the fix

The weakness sits in the parts of MONAI Label that store uploaded images and labels: a file identifier supplied by the caller is appended to the storage folder and written to disk without checking that the result stays inside it. The server writes with the privileges of the account it runs under. Authentication is off by default, so any client that can reach the server over the network can trigger the write. Versions 0.8.5 and earlier are affected. MONAI Label is a separate program from the wider MONAI framework.

Maintainers confirmed the flaw and merged a source fix on 19 October that forces uploads to resolve inside approved folders and rejects empty names or names beginning with a dot. Their advisory, published in November, still lists no patched version and says the correction will come in a future release; the newest release listing remains 0.8.5, which dates to 2024. The CVE record, CVE-2026-133968, rates the flaw 9.8 out of 10. The project's advisory page still shows no CVE, which The Hacker News describes as an updating lag rather than a second flaw.

Attribution and what is not established

Amazon attributes the campaign to an unidentified China-linked actor with moderate confidence, citing Simplified Chinese prompts, China Standard Time settings and targets likely to be of strategic interest to China. The company did not connect the work to a named group or establish that a government was behind it.

Root access was obtained on one university server. Whether another MONAI Label deployment would give the same result depends on how it is run and what its service account can do, and the campaign also included failed operations. Amazon's report does not show that patient scans or records were taken, that clinical systems were touched, or that the attacker reached beyond the university targets described.

What is not known yet

  • Who the operator is: Amazon did not name a group or establish government involvement.
  • Whether the second university network was actually compromised, which Amazon infers but has not confirmed.
  • Whether any patient scans, images or records were accessed or taken.
  • How many MONAI Label servers are reachable from the internet with authentication disabled.
  • When a patched release will be published, since only a source-code fix exists.

What to do

If you run MONAI Label

  1. Block untrusted network access to the server with firewall and routing rules rather than relying on hard-to-guess URLs.
  2. Turn on authentication, which the server leaves off by default.
  3. Run the server under a limited account instead of root and restrict the folders it can write to.
  4. Check reachable servers for unexpected Python startup files and unusual training jobs, and confirm any source build includes the October fix before treating it as patched.

If you operate AI agent tooling

  1. Put authentication on any Cairn or similar agent control panel and keep it off the public internet.
  2. Review what the agents' credentials and network position allow them to reach, since exposed run logs revealed the whole campaign.

For everyone else

  1. Treat software as unpatched when a project says a fix is merged but no installable release exists.
  2. Review internet-facing tools that ship with authentication disabled by default and close or gate them.

Update log

  • First published.

Sources

New incident reports by email

One email a week with the reports published since the last one.

We email you about this topic only; unsubscribe any time. See our Privacy Policy.

Want us to look at your domain?

Give us one apex domain and we'll send back a written snapshot of what is already visible from outside. No call required.