AWS Describes Cairn Agents’ Exploitation of a Zero-Day Flaw in MONAI Label
AWS said Cairn framework agents discovered and exploited a critical path traversal flaw in the MONAI Label application during a September campaign. Versions up to and including 0.8.5 are vulnerable.

Cairn in MONAI Label reportedly demonstrated how a multi-agent tool can independently find and exploit a previously unknown flaw after being given a general objective. On November 20, AWS published findings from a campaign conducted between September 8 and 10, 2026, during which it found a publicly accessible, unauthenticated Cairn instance containing an attacker’s operational history.
The target was the open-source MONAI Label application, designed for working with medical images. The MONAI project confirmed a critical path traversal vulnerability affecting versions up to and including 0.8.5. With authentication disabled, as in the default configuration, the flaw may allow arbitrary file writes and subsequent remote code execution.
Cairn in MONAI Label Obtained File Write Access and a Root Shell
According to AWS, the agents were given the goal “obtain shell access.” They then reportedly discovered the path traversal flaw and, in approximately 11 minutes, gained the ability to write arbitrary files to the server. In a later session, they reportedly exploited the same flaw to write a reverse shell into the sitecustomize.py file, thereby obtaining a root shell.
The description does not mean the attack was entirely free of human intervention. The agents independently planned and carried out steps only after a human gave them the objective. After the vulnerability was discovered, the operator reportedly manually provided a proxy for access to another network, according to AWS.
The case therefore demonstrates an agent-based tool’s ability to automate attack steps from vulnerability discovery through exploitation. However, the available information does not show that Cairn reliably or generally automates zero-day vulnerability discovery without human guidance.
Critical Flaw Without a Designated Released Fixed Version
The vulnerability is tracked as CVE-2026-133968. On November 20, the NVD and GitHub Advisory Database published it with a CVSS score of 9.8 and stated that it had been exploited in the wild since September 2026.
The MONAI project’s security advisory identifies a source-code fix through pull request #1915, but does not yet designate a released fixed version. MONAI Label operators should therefore check whether they are using a version up to and including 0.8.5 and apply the available source-code fix or recommended mitigations.
- Enable authentication.
- Do not expose the server to untrusted networks.
- Run the service without privileges.
- Check instances for possible signs of compromise, especially if publicly accessible.
AWS did not state which universities may have been attacked, nor did it publish independently verifiable forensic data; the scope of the compromise therefore cannot be independently quantified.
Attribution Remains AWS’s Assessment
AWS attributes the activity to an actor with a possible connection to China, with a moderate level of confidence. The assessment is based on Chinese prompts, the time zone, and target selection. It is not a confirmed attacker identity or evidence of state direction.
Further developments will depend mainly on the release of a new MONAI Label version with a clearly identified fix and supported versions. Any inclusion of CVE-2026-133968 in CISA’s Known Exploited Vulnerabilities catalog should also be monitored, along with independent confirmation of the incident’s scope or attribution.
Sources
- AWS Security Blog – AWS’s primary findings on the campaign, the agents’ autonomous steps, the root shell, the timeline, and the assessment of a connection to China.
- Project MONAI / GitHub Security Advisory GHSA-qc22-rmgc-vg94 – Confirms the flaw in MONAI Label, affected versions, the absence of a released fixed version, the source-code fix, and mitigations.
- GitHub Advisory Database / NVD – Confirms CVE-2026-133968, the critical score of 9.8, and the claim of exploitation since September 2026.
- BleepingComputer – Independently summarizes AWS’s disclosure and the identification of the vulnerable application as MONAI Label.
Verified and updated: November 21, 2026 06:20


