Security Guides

MONAI Label Critical Path Traversal Flaw (No CVE, 0.8.5 and Earlier) Reported Exploited With an Autonomous AI Tool: What Server Operators Should Do

MONAI Label 0.8.5 and earlier has a critical path traversal flaw allowing arbitrary file writes; auth is off by default. AI-found attacks reported.

Published 2026-11-21 Updated 2026-11-21 Last verified 2026-11-21 8 min read
On this page

For: administrators, researchers and developers who run a MONAI Label server (the open-source medical-image annotation tool in the MONAI project), especially at universities and hospitals. This article is based on the project's security advisory and BleepingComputer's report, and does not cover attack techniques.

Developing: this article will be updated as the project publishes more

Who is affected

administrators, researchers and developers who run a MONAI Label server (the open-source medical-image annotation tool in the MONAI project), especially at universities and hospitals.

Confirmed so farSee the table ↓
  • Affected versions
  • Severity
  • Authentication
1 more
  • Fix in code

Listed as possibly affected

  • Attacks in the wild
  • Use of an autonomous AI tool to find the flaw
  • Targets
Not yet confirmed
  • A released version that contains the fix
  • A CVE identifier
  • The number of affected servers
1 more
  • Whether the project confirms the reported attacks
What to do nowSee the steps ↓
  1. Find every MONAI Label server and check its version and exposure
  2. Take it off untrusted networks
  3. Turn authentication on
3 more
  1. Limit what the server process can write
  2. Check servers that were reachable from outside
  3. Update when a release with the fix appears
Last checked
As of 21 November 2026
Follow-up plans and what we checked

The advisory says the fix will be in the next release. As of 21 November 2026 the newest release on PyPI is 0.8.5 and the advisory lists no patched version.

What MONAI Label operators should do today

1

Find every MONAI Label server and check its version and exposure

List the servers you run (including lab and teaching machines), note whether the version is 0.8.5 or earlier, and check from outside your network whether the port answers. A research server that was opened to the internet for convenience is the highest-priority case.

2

Take it off untrusted networks

The advisory's own workaround is not to expose the server to untrusted networks. Put it behind your institution's VPN or an internal-only address, and restrict access to the people who need it. A documentation change on 19 November 2026 in the project's repository also adds a warning against internet-facing deployments.

3

Turn authentication on

The advisory says authentication is disabled by default and gives the setting to enable it (MONAI_LABEL_AUTH_ENABLE=True). Even with authentication on, the annotator role is enough to use the flaw, so give that role only to people you trust. See multi-factor authentication for the wider habit.

4

Limit what the server process can write

The advisory suggests running the server as an unprivileged user whose write access is limited to the datastore directory. This reduces what a file-write flaw can reach; it does not remove the flaw.

5

Check servers that were reachable from outside

If a vulnerable server was reachable from the internet, look for files you did not put there, unfamiliar accounts or processes, and outbound connections you cannot explain. If you find any, treat the host as compromised: isolate it and rebuild it rather than cleaning in place. Servers on the same network may need review too, since the press report says the operators reportedly gave the agent access to a second university network running the same software and pointed it at other machines (outcome not reported).

6

Update when a release with the fix appears

The fix is merged into the project's main branch (pull request #1915) and the advisory says it will ship in the next release. Until then, avoid running unreviewed code from the main branch in production unless your team has reviewed it. Check the releases page, and update as soon as a release includes the fix.

What happened (the advisory and the press report)

  1. 8 to 10 September 2026

    Per BleepingComputer's report of a cloud provider's threat intelligence findings, attackers used an autonomous AI penetration-testing platform against servers, mainly at education institutions in East Asia, and found and used this flaw.
  2. 19 October 2026

    The fix (pull request #1915) is merged into the project's main branch.
  3. 18 November 2026

    The project publishes the advisory (GHSA-qc22-rmgc-vg94), rated Critical, for versions 0.8.5 and earlier. It lists no patched version.
  4. 20 November 2026

    BleepingComputer reports the observed attacks.
What the advisory says
Product
MONAI Label (pip package monailabel), an open-source medical-image annotation server
Identifier
GHSA-qc22-rmgc-vg94; no CVE assigned
Severity and weakness
Critical; CWE-22, improper limitation of a pathname to a restricted directory
Affected
Version 0.8.5 and earlier
Flaw
The upload endpoints for images and labels use a caller-supplied ID without validating it, and the datastore does not confirm the resulting path stays inside its directory, so uploaded files can be written outside the datastore wherever the server process can write. The advisory says this could lead to remote code execution or denial of service.
Who can use it
Authentication is disabled by default, so any network client that can reach the server can; with authentication enabled, the annotator role is sufficient
Fix
Pull request #1915 makes both upload paths reject locations outside the datastore directories, and makes the image endpoint reduce the ID to a file name and reject empty or dot-prefixed IDs. The advisory lists no patched version; it is to be in the next release.
Reported by
A reporter credited in the advisory, after an original private report from a cloud provider's vulnerability-disclosure team

What is known and what is not

The state as of 21 November 2026. When the project publishes more, the status column will be rewritten.

ItemWhat the source saysStatus
Affected versions0.8.5 and earlier (advisory)Confirmed (project advisory)
SeverityCritical (advisory)Confirmed (project advisory)
CVENone known (advisory)Not stated
AuthenticationDisabled by default (advisory)Confirmed (project advisory)
Fix in codeMerged to the main branch on 19 October 2026 (pull request #1915)Confirmed (project repository)
Fixed releaseNone listed in the advisory; newest PyPI release is 0.8.5Under investigation
Press says a security update was released 18 NovemberWe could not find a matching releaseNot stated
Attacks in the wildObserved 8 to 10 September 2026 by a cloud provider's threat intelligence team (BleepingComputer)Possible (press report)
Use of an autonomous AI tool to find the flawReported by BleepingComputer, citing the providerPossible (press report)
TargetsMainly education institutions in East Asia (press)Possible (press report)
Number of affected serversNot disclosedNot stated
Exploitation mentioned in the advisoryNot mentionedNot stated

How to read this: the advisory and the press report are separate sources

The advisory is the project's own statement about the flaw and the fix. The reports of attacks and of an AI tool come from a press article summarising a cloud provider's findings; we did not see the provider's report itself. The press article and the advisory differ on whether a security update has been released, so check the project's releases page directly.

What this flaw can lead to

In general, a server that accepts files and lets the sender influence where they are saved can be turned into a way to place code on that server. This is why the advisory mentions remote code execution. It does not say how, and this article does not either.

  • An unauthenticated, internet-reachable annotation server: a stranger can reach the flaw without any account
  • A research server on a university network: a compromise may give access to other hosts on the same network
  • A server holding medical images: the data on it may be sensitive, so a compromise has privacy consequences beyond the host

Sources (public records)

The facts in this article come from the public sources below. Undisclosed details are not speculated on.

Update history

2026-11-21: First version, based on the project's advisory (18 November 2026) and BleepingComputer's report. No fixed release was found; this article will be updated when one is published or a CVE is assigned.

Latest breaking reports

All breaking reports →

FAQ

QWhich MONAI Label versions are affected?
A

The project's advisory (GHSA-qc22-rmgc-vg94) says versions 0.8.5 and earlier of the monailabel package are affected. It lists no patched version. The fix is in a pull request that the project says will be included in the next release.

QIs there a fixed release?
A

As of 21 November 2026 we found none. The advisory lists no patched version, and the newest release on the Python Package Index (PyPI) is still 0.8.5, uploaded in November 2024. The fix (pull request #1915) is merged into the project's main branch. BleepingComputer says the maintainers released a security update on 18 November; we could not find a release matching that, so check the project's releases page before relying on either statement.

QIs there a CVE?
A

No. The advisory says there is no known CVE. It is rated Critical, and the weakness is classified as CWE-22 (path traversal).

QWho is at risk?
A

Anyone running MONAI Label 0.8.5 or earlier where a client that should not have access can reach the server. Authentication is disabled by default, so in the default configuration any network client that can reach it can use the flaw. With authentication enabled, the advisory says the annotator role is sufficient.

QWere attacks seen in the wild?
A

BleepingComputer reports that a cloud provider's threat intelligence team observed attackers exploiting it on 8 to 10 September 2026, mainly against education institutions in East Asia, using an autonomous AI penetration-testing platform. That is a press report of the provider's findings; the project's advisory does not mention exploitation.

QWhat should I do if I cannot update yet?
A

The advisory lists three workarounds: enable authentication and do not give the annotator role to untrusted users; do not expose the server to untrusted networks; and run the server as an unprivileged user whose write access is limited to the datastore directory. Also look for unexpected files and sessions on any server that was reachable from the internet.