CVE-2026-100840: MONAI: code execution via malicious bundle config
HIGHPoC AVAILABLECISA: ATTENDMONAI through 1.6.0 resolves `_target_` values in bundle configurations to any importable callable and passes `$` expressions to Python `eval()`. A malicious bundle therefore runs arbitrary code as soon as someone loads it with `monai.bundle.load()` or `monai.bundle.run()`. The CVSS is 7.8 (local, user interaction required), the EPSS is low at 0.00215 (89th percentile), it is not in CISA KEV, and there is no public exploit or Nuclei template. CISA SSVC rates it ATTEND, so it deserves scheduled attention rather than an emergency response. The real risk is supply chain: medical imaging teams often pull pretrained bundles from the MONAI Model Zoo, Hugging Face or colleagues, and those workstations and GPU servers often hold sensitive data. Until a fixed release is available, only load bundles from sources you trust and have reviewed, and run bundle loading in isolated, least-privilege environments. Follow GHSA-873f-pvrv-4x83 for the patch.
What is the risk?
Overall risk is MEDIUM-HIGH for organizations that load third-party MONAI bundles, and LOW for those that use only internally built bundles. The code is trivial to exploit once a victim loads a crafted bundle: attack complexity is low and no privileges are needed. Exploitation does require user interaction (loading the bundle) and local access, which limits mass exploitation. Impact on confidentiality, integrity and availability is high, because the code runs with the privileges of the user or service loading the bundle. EPSS is low (0.00215), there is no KEV listing and no public exploit, so active exploitation is not currently observed. Exposure is higher in healthcare and research settings, where unvetted models are routinely shared.
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) Monitor the GitHub advisory GHSA-873f-pvrv-4x83 and the VulnCheck advisory, and upgrade MONAI as soon as a release newer than 1.6.0 with a fix is available. 2) Until then, load only bundles from trusted, verified sources; inspect the config files (JSON/YAML) for
_target_entries that reference unexpected modules and for$expressions before loading. 3) Runmonai.bundle.load()andmonai.bundle.run()in sandboxed containers or VMs with no credentials, restricted network egress and read-only data mounts. 4) Do not run bundle loading as root or with access to production secrets or PHI. 5) Pin bundle sources and verify hashes or signatures in CI/CD and MLOps pipelines. 6) Detection: alert on unexpected child processes, outbound connections or file writes from Python processes running MONAI, and audit which bundles were loaded from external sources.
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-2026-100840?
MONAI through 1.6.0 resolves `_target_` values in bundle configurations to any importable callable and passes `$` expressions to Python `eval()`. A malicious bundle therefore runs arbitrary code as soon as someone loads it with `monai.bundle.load()` or `monai.bundle.run()`. The CVSS is 7.8 (local, user interaction required), the EPSS is low at 0.00215 (89th percentile), it is not in CISA KEV, and there is no public exploit or Nuclei template. CISA SSVC rates it ATTEND, so it deserves scheduled attention rather than an emergency response. The real risk is supply chain: medical imaging teams often pull pretrained bundles from the MONAI Model Zoo, Hugging Face or colleagues, and those workstations and GPU servers often hold sensitive data. Until a fixed release is available, only load bundles from sources you trust and have reviewed, and run bundle loading in isolated, least-privilege environments. Follow GHSA-873f-pvrv-4x83 for the patch.
Is CVE-2026-100840 actively exploited?
Proof-of-concept exploit code is publicly available for CVE-2026-100840, increasing the risk of exploitation.
How to fix CVE-2026-100840?
1) Monitor the GitHub advisory GHSA-873f-pvrv-4x83 and the VulnCheck advisory, and upgrade MONAI as soon as a release newer than 1.6.0 with a fix is available. 2) Until then, load only bundles from trusted, verified sources; inspect the config files (JSON/YAML) for `_target_` entries that reference unexpected modules and for `$` expressions before loading. 3) Run `monai.bundle.load()` and `monai.bundle.run()` in sandboxed containers or VMs with no credentials, restricted network egress and read-only data mounts. 4) Do not run bundle loading as root or with access to production secrets or PHI. 5) Pin bundle sources and verify hashes or signatures in CI/CD and MLOps pipelines. 6) Detection: alert on unexpected child processes, outbound connections or file writes from Python processes running MONAI, and audit which bundles were loaded from external sources.
What systems are affected by CVE-2026-100840?
This vulnerability affects the following AI/ML architecture patterns: training pipelines, model serving, medical imaging workflows, MLOps pipelines, Jupyter notebooks.
What is the CVSS score for CVE-2026-100840?
CVE-2026-100840 has a CVSS v3.1 base score of 7.8 (HIGH). The EPSS exploitation probability is 0.31%.
What is the AI security impact?
Affected AI Architectures
MITRE ATLAS Techniques
AML.T0010.003ModelAML.T0011.000Unsafe AI ArtifactsAML.T0050Command and Scripting InterpreterAML.T0058Publish Poisoned ModelsAML.T0112.001AI ArtifactsCompliance Controls Affected
What are the technical details?
Original Advisory
MONAI through 1.6.0 contains a remote code execution vulnerability in the bundle configuration engine that resolves _target_ values to arbitrary importable callables without an allow list and passes $ expressions to Python eval(). Attackers can publish a malicious bundle with crafted configuration containing arbitrary code that executes when a victim loads the bundle using monai.bundle.load() or monai.bundle.run().
Exploitation Scenario
An attacker publishes a plausible-looking MONAI bundle (for example a segmentation model for a popular imaging task) to a public repository, model hub or shared drive. The `configs/*.json` file contains a `_target_` pointing to a callable such as `os.system` or `subprocess.run`, or a `$` expression that calls `__import__('os').system(...)`. A researcher or ML engineer downloads the bundle and runs `monai.bundle.load()` or `monai.bundle.run()` to test it. The payload executes with that user's privileges, steals cloud credentials, SSH keys and access to imaging datasets, and installs a reverse shell on the GPU workstation. The attacker then moves laterally into the research or hospital network.
Weaknesses (CWE)
CWE-95Improper Neutralization of Directives in Dynamically Evaluated CodePrimaryCWE-95Improper Neutralization of Directives in Dynamically Evaluated CodeCWE-95 — Improper Neutralization of Directives in Dynamically Evaluated Code: The product receives input from an upstream component, but it does not neutralize or incorrectly neutralizes code syntax before using the input in a dynamic evaluation call (e.g. "eval").
- [Architecture and Design, Implementation] If possible, refactor your code so that it does not need to use eval() at all.
- [Implementation] Assume all input is malicious. Use an "accept known good" input validation strategy, i.e., use a list of acceptable inputs that strictly conform to specifications. Reject any input that does not strictly conform to specifications, or transform it into something that does. When performing input validation, consider all potentially relevant properties, including length, type of input, the full range of acceptable values, missing or extra inputs, syntax, consistency across related fields, and conformance to business rules. As an example of business rule logic, "boat" may be syntactically valid because it only contains alphanumeric characters, but it is not valid if the input is only expected to contain colors such as "red" or "blue." Do not rely exclusively on looking for malicious or malformed inputs. This is likely to miss at least one undesirable input, especially if the code's environment changes. This can give attackers enough room to bypass the intended validation. However, denylis
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