Skip to content
← Blog · · df00tech

CVE-2026-48705: Malformed DICOM Value Representations Crash or Silently Corrupt MONAILabel Inference

vuln-intel pip CVE-2026-48705 #checker-reject

What's Happening

CVE-2026-48705 is an input-validation flaw (CWE-20) in MONAI's DICOM handling. Crafted DICOM files containing malformed DS (Decimal String) or IS (Integer String) value representations can bypass pydicom's validation logic before reaching the MONAILabel inference pipeline. Depending on how the malformed values propagate, the result is either a hard crash of the MONAILabel inference server — a denial-of-service condition — or, more insidiously, silent corruption of the tensor inputs built from the DICOM data. The latter case is the more dangerous outcome: the server keeps running and returns an inference result, but that result was computed from corrupted input, meaning a medical imaging pipeline can produce incorrect diagnostic output without any visible error.

Affected Software

The vulnerability affects the monai package (distributed via pip) in all versions prior to 1.6.1. No patched release date has been published yet.

Exploitation Status

A public proof-of-concept exists for this vulnerability. It has not been confirmed as weaponized in active campaigns or added to CISA's KEV catalog, but the public PoC lowers the bar for exploitation significantly. With a CVSS score of 9.3, and given that MONAILabel is frequently deployed in clinical and research imaging environments, defenders running MONAI-based inference servers should treat this as an urgent patching and monitoring priority — particularly where inference endpoints accept DICOM uploads from less-trusted sources.

How Our Detection Catches It

Our detection ships across Microsoft Sentinel (KQL), Splunk (SPL), Elastic (EQL), QRadar (AQL), Sumo Logic, Chronicle (YARA-L), and CrowdStrike (CQL), and layers several signals:

  • Process-level crash detection — identifies MONAILabel/MONAI server process terminations and crash signatures consistent with unhandled parsing exceptions.
  • Application log analysis — flags Python tracebacks that originate specifically in pydicom's valuerep module or other DICOM-parsing code paths during active inference requests, distinguishing this failure mode from generic server errors.
  • Upload anomaly detection — surfaces anomalous or malformed DICOM file uploads directed at inference endpoints, which can indicate exploitation attempts even before a crash occurs.
  • Environment exposure checks — identifies hosts running vulnerable monai package versions (< 1.6.1), supporting proactive patch prioritization independent of runtime activity.

Together these signals give defenders visibility into both the loud failure mode (server crash) and the quiet one (corrupted inference without a crash), which is the harder case to catch without this kind of targeted monitoring.

Full Detection Details

For the complete detection logic, platform-specific queries, and investigation guidance, see the full detection page for CVE-2026-48705.

Get new detections in your inbox

New ATT&CK coverage plus CISA KEV / CVE detection rules, roughly weekly. No spam, unsubscribe anytime.