MONAI Label at a glance
Amazon Threat Intelligence revealed on November 20, 2026 that Cairn AI agents from a China-linked actor found and exploited on their own a zero-day in MONAI Label (CVE-2026-133968, CVSS 9.8) to root access on a university server; there is no patched release.
Published November 20, 2026. Organization: Amazon. Category: Security. Status: Workaround available. Relevance: Important. Last verified: November 20, 2026. Products: MONAI Label, Cairn.
Status
- Status: Workaround available
- Severity: Critical
- Affected: MONAI Label 0.8.5 and earlier (pip package
monailabel): a caller-controlledimageparameter inPUT /datastore/image, and the same pattern inPUT /datastore/label, allows arbitrary file writes and remote code execution - Installations with authentication disabled, the default (
MONAI_LABEL_AUTH_ENABLE=False), where any network client can exploit the flaw; with authentication enabled, the ANNOTATOR role is enough - Two East Asian university networks attacked September 8-10, 2026; one server was taken with root access and the same flaw was inferred in the second network
- CVE identifiers: CVE-2026-133968 (NVD)
- Last reviewed: November 20
- Detected by: Autonomous Cairn agents operated by a suspected China-linked actor; disclosed by Amazon Threat Intelligence. Detected by an AI system.
What changed with MONAI Label
Amazon Threat Intelligence revealed on November 20, 2026 that autonomous AI agents operating with Cairn, an open penetration-testing framework, found and exploited a zero-day in MONAI Label, an open AI application for medical imaging. Amazon found the actor's Cairn server, exposed without authentication, on September 10, 2026 and reconstructed a roughly 48-hour campaign from September 8, 2026 against East Asian universities outside mainland China: with the goal 'obtain shell access', the agents mapped the MONAI Label API, probed its upload endpoint and went from mapping to arbitrary file writes in about 11 minutes; they then wrote a Python reverse shell into sitecustomize.py and triggered a training endpoint to run it as root.
The flaw, CVE-2026-133968, is a path traversal in the PUT /datastore/image and PUT /datastore/label endpoints of MONAI Label: the National Vulnerability Database entry published on November 20, 2026 rates it 9.8 out of 10 (critical, CVSS v3.1) and lists versions 0.8.5 and earlier as affected, with authentication disabled by default. The agents only succeeded there: one operation ran 148 actions over about 16 hours against a Redis server without success, another tried more than 30,000 credential combinations against a university GPU server, and the actor later reused the zero-day through a proxy tunnel to a second university network. Project MONAI merged the fix (pull request #1915) on October 19, 2026 and published advisory GHSA-qc22-rmgc-vg94 on November 18, 2026, but as of November 20, 2026 there is no patched release.
How to fix: MONAI Label
- Workaround — Enable authentication (
MONAI_LABEL_AUTH_ENABLE=True) and do not grant the ANNOTATOR role to untrusted users. (GHSA-qc22-rmgc-vg94) - Workaround — Do not expose the MONAI Label server to untrusted networks until a patched release is installed. (GHSA-qc22-rmgc-vg94)
- Configuration — Run the server as an unprivileged user with write access limited to the datastore directory. (GHSA-qc22-rmgc-vg94)
- Update — Apply the code fix merged in pull request #1915 when the next MONAI Label release is published; as of November 20, 2026 there is no patched release. (GHSA-qc22-rmgc-vg94)
- Monitor — Prioritize high-confidence detections that trigger an automated response, with people overseeing outcomes instead of reviewing every individual action (Amazon Threat Intelligence guidance). (Amazon Threat Intelligence)
MONAI Label in numbers
The values come from the source linked on the date shown. They are not IA Compa measurements.
- 9.8 — CVSS v3.1 base score for CVE-2026-133968 (NVD), as of Nov 20, 2026. Source: nvd.nist.gov
- 11 — Minutes from mapping the API to arbitrary file writes (Amazon), Cairn, as of Nov 20, 2026. Source: aws.amazon.com
- 148 — Agent actions in the failed operation against Redis (Amazon), Cairn, as of Nov 20, 2026. Source: aws.amazon.com
- 30,000 — Credential combinations tried against a GPU server (Amazon), Cairn, as of Nov 20, 2026. Source: aws.amazon.com
Why MONAI Label matters for businesses
Why it matters: offensive agents compress to minutes the time between a flaw existing and a server falling, while many businesses still patch on weekly or monthly cycles, so any business running AI servers, imaging tools or agent frameworks with default settings is exposed in that gap. A business can use the same approach defensively: run agent-driven penetration tests against your own systems before someone else does, enable authentication and least-privilege access on every AI tool you host, and automate first response actions so people review outcomes instead of triaging every alert by hand.
Sources
- Chinese-nexus actor's Cairn AI agents found a zero-day on their own — Amazon Threat Intelligence, aws.amazon.com — Official
- GHSA-qc22-rmgc-vg94: Path traversal in datastore upload endpoints allows arbitrary file write — Project MONAI, github.com — Official
- CVE-2026-133968 — National Vulnerability Database, nvd.nist.gov — Official
Last verified: November 20, 2026. IA Compa summarizes public announcements for business readers. We have no relationship with the companies mentioned. Always confirm details, prices and availability with the official source.
