Skip to content
TrustList
Security advisories

MONAI Label zero-day found by attackers' AI agents has no patched release yet

Editorial

By TrustList Editorial

A path traversal in MONAI Label 0.8.5 and earlier lets anyone who can reach the server write files anywhere; Amazon says attackers' AI agents found and used it in about 11 minutes, and no fixed release is out.

About MONAI Label zero-day found by attackers' AI agents has no patched release yet

MONAI Label zero-day found by attackers' AI agents has no patched release yet

20 November 2026: MONAI Label, an open-source tool for labelling medical images with AI, has a critical path traversal flaw that lets anyone who can reach the server write files anywhere the process can write. The project's advisory, GHSA-qc22-rmgc-vg94, covers every release up to 0.8.5 and lists no patched version: the fix is merged in the code but will only ship in the next release. Amazon says the flaw was first found and used by an attacker's AI agents, not by a person.

Not yet independently verified. The flaw, affected versions, workarounds and the missing release come from the project's GitHub advisory (published 18 November 2026) and PyPI, where 0.8.5 of 25 November 2024 is still the latest release (both read on 21 November 2026). The campaign details are Amazon's own account; Amazon rates the zero-day finding as high confidence and the China link as moderate. No CVE had been assigned when we checked. We will update this when it can be confirmed, and remove this note.

How the flaw works

The PUT /datastore/image endpoint takes an image ID from the caller and uses it, unchecked, to build the path where the uploaded file is saved. An ID such as ../../../tmp/x writes the upload outside the datastore. The label endpoint, PUT /datastore/label, builds its path the same way. The advisory says this can lead to code execution, for example by overwriting application code or adding an SSH key or a cron entry, or to a denial of service by overwriting configuration.

Authentication is off by default in MONAI Label, so in the default setup any network client can do this. With authentication on, the ANNOTATOR role is enough.

What Amazon saw

Amazon Threat Intelligence says that on 10 September 2026 it found a publicly reachable, unauthenticated server running Cairn, an open-source framework for AI penetration testing. Its records showed a campaign of about 48 hours from 8 September against universities in East Asia outside mainland China. Told to get shell access, the agents found a MONAI Label upload endpoint, tested it for path traversal and, in about 11 minutes, went from mapping the API to writing files on the server. Amazon says the attacker then used the same flaw on a second university network. Other attempts failed, including more than 30,000 password guesses against SSH and other services. Amazon assesses with moderate confidence that the actor is linked to China, and with high confidence that the flaw was a zero-day, as it found no earlier fix, issue or advisory.

Steps for teams running MONAI Label

  • Find every MONAI Label server, in research groups and hospitals as well as IT-run systems, and take any that face untrusted networks off them.
  • Turn on authentication (MONAI_LABEL_AUTH_ENABLE=True) and do not give the ANNOTATOR role to untrusted users.
  • Run the server as an unprivileged user that can write only to the datastore directory, as the advisory advises.
  • Watch the project for the release that contains pull request 1915, and look for files written outside the datastore since September.

Sources

Categories & features

TrustList Weekly

The week in software and IT, in one email

The news that matters to buyers, new rankings and our own research. Every Thursday, free, and easy to leave.

We will email you to confirm. Unsubscribe with one click in any issue. Privacy policy

You may be interested in