Zero Radar
← feed
CriticalVulnerabilityCVE-2026-48705updated 27 Oct 2026, 13:30 UTC

MONAILabel medical imaging server can be crashed or corrupted by one crafted file

Anyone who can reach the server can upload a single specially crafted medical scan file. This can make the server crash or quietly produce wrong AI results on patient images, with no login needed.

Fixed in monai 1.6.1Official source
SeverityCriticalCVSS 9.3 of 10
Being exploited?UnclearNo reports either way yet
Fix available?Yesmonai 1.6.1 · since 27 Oct 2026
How sure is this?Official sourceBacked by an official source

How the attack works simplified

  1. 1AttackerHas network access to the server
  2. 2Crafted DICOM uploadSingle ~132 KB file via unauthenticated PUT
  3. 3Invalid NaN valuespydicom accepts NaN/inf in numeric fields
  4. 4Affine matrix poisonedMONAI skips finite checks on spacing
  5. 5Crash or wrong resultsInference pipeline fails or silently corrupts output
  6. ResultServer crashes or AI results are corrupted

What happened

pydicom accepts invalid NaN and infinity values in DICOM numeric fields; MONAI's PydicomReader builds spatial transforms from them without checking, and MONAILabel's default REST API accepts uploads without authentication.

Why it is dangerous: Crashes or silently wrong segmentation results can affect clinical AI workflows; the attacker needs only network access to the default port 8000.

Are you affected?

You are affected if you run MONAILabel (pip package monai below 1.6.1) with its default unauthenticated REST API reachable over the network.

Package / productAffected versionsFixed in
monai · pip< 1.6.11.6.1

What to do

  1. Upgrade monai to 1.6.1 or later (pip install -U monai>=1.6.1)
  2. Restrict network access to the MONAILabel server (port 8000) and do not expose it publicly
  3. Enable authentication on the MONAILabel REST API where possible

Timeline

  1. 27 Oct 2026, 13:30 UTC
    Publicly disclosed
  2. 27 Oct 2026, 13:30 UTC
    Reported by GitHub Advisory · GitHub Advisory
  3. 27 Oct 2026, 13:30 UTC
    Picked up by Zero Radar

Sources 1 report, oldest first

Official vendor, government CERT or advisory database

Technical details

ID
CVE-2026-48705 · GHSA-6hp3-vr39-rqw8
CVSS
9.3 / 10
Disclosed
27 Oct 2026, 13:30 UTC
On Zero Radar
27 Oct 2026, 13:30 UTC (0 min after disclosure)
Last change
27 Oct 2026, 13:30 UTC

The plain-language parts and the attack diagram were written by an AI model (anthropic/claude-haiku-5.5) from the sources above. Facts like versions, exploitation and dates come from the sources themselves. How Zero Radar works

MONAILabel medical imaging server can be crashed or corrupted by one crafted file · Zero Radar