Overview
CVE-2026-48705 affects MONAI (Medical Open Network for AI), an open source AI toolkit for healthcare imaging. It was published on November 5, 2026 through GitHub advisory GHSA-6hp3-vr39-rqw8 with a CVSS 3.1 score of 9.3 (vector AV:N/AC:L/PR:N/UI:N/S:C/C:N/I:H/A:L). The weaknesses are CWE-20 (improper input validation) and CWE-682 (incorrect calculation).
Affected Versions
| Product | Affected | Fixed |
|---|---|---|
| MONAI | All versions before 1.6.1 | 1.6.1 |
Technical Summary
PydicomReader._get_affine() trusts non-finite values in three DICOM fields: PixelSpacing, ImagePositionPatient, and ImageOrientationPatient. It composes them into an affine matrix without checking that they are finite numbers.
An unauthenticated attacker who can upload a crafted DICOM file to a default MONAILabel deployment can push NaN or infinity values through loading, spacing, inference, and output. Possible results:
- Inference failures that disrupt the service.
- Poisoned datastore records that other users or training jobs later consume.
- Geometrically corrupted outputs, such as segmentations that no longer line up with the source scan.
The integrity impact is the main concern. Corrupted segmentation or measurements in a clinical or research setting can go unnoticed. The scope-changed score reflects that bad data leaves the vulnerable component and affects downstream systems.
Remediation
- Upgrade MONAI to 1.6.1 or later.
- Do not expose MONAILabel to untrusted networks. Put it behind authentication or a VPN.
- Validate uploaded DICOM files and reject non-finite spacing, position, and orientation values before they reach the reader.
- Audit datastore records created from untrusted uploads and remove entries with NaN or infinite geometry.
References
- NVD: CVE-2026-48705
- GitHub Security Advisory GHSA-6hp3-vr39-rqw8