CVE-2026-100842: MONAI: eval sandbox bypass allows code execution

HIGHCISA: ATTEND
Published September 27, 2026
CISO Take

MONAI through 1.6.0 validates shape expressions in its bundle verification flow with an AST allowlist that only inspects variable names, so expressions built from constants, attributes and calls (for example object-introspection chains) slip through and reach eval(), giving code execution. The flaw is CVSS 8.8 but requires local access with low privileges, and it only triggers in the non-default 'verify_net_in_out' bundle flow when an attacker can influence a bundle's metadata. There is no public exploit, no Nuclei template and it is not in CISA KEV; however, EPSS sits at the 97th percentile (score 0.00148), CISA SSVC says ATTEND, and 126 downstream packages depend on MONAI, so the realistic risk is a poisoned or untrusted bundle (model zoo download, shared pipeline) executing code on a researcher workstation or training host. Treat bundles from untrusted sources as executable code, avoid running 'verify_net_in_out' on them until a fixed release is available (no patched version is listed yet; track GHSA-h89g-r5pc-wxfm), and inventory where MONAI bundles are pulled and verified in your imaging and clinical AI pipelines.

Sources:NVDEPSSGitHub Advisoryvulncheck.comATLASOpenSSF

What is the risk?

Medium-high overall. Technical severity is high (CVSS 8.8, scope changed, full CIA impact) and the bypass is trivial once an attacker controls bundle metadata, but exploitation requires local execution of a non-default CLI flow and attacker influence over bundle content. No public exploit, not in KEV, low absolute EPSS (0.00148) though ranked in the top 97% percentile, SSVC ATTEND. The main exposure is supply chain: bundles downloaded from model hubs or exchanged between institutions. Package risk score is 26/100, OpenSSF Scorecard 7.9/10, and the package has 14 other CVEs, indicating a recurring vulnerability history.

How does the attack unfold?

Initial Access
Attacker distributes a MONAI bundle whose metadata contains a crafted shape expression via a model hub or shared repository.
AML.T0010.003
Execution
A user or pipeline runs the 'verify_net_in_out' flow, and the AST allowlist (checking only ast.Name nodes) passes the expression to eval().
AML.T0011.000
Sandbox Escape
Object introspection chains reach builtins or os modules, giving arbitrary Python code execution.
AML.T0050
Impact
Attacker steals credentials, datasets and model weights from the host and may pivot to other systems.
AML.T0025

What systems are affected?

PackageEcosystemVulnerable RangePatched
MONAIpip—No patch
8.8KOpenSSF 7.8129 dependentsPushed 4d ago41% patched~22d to patchFull package profile →

Do you use MONAI? You're affected.

How severe is it?

CVSS 3.1
8.8 / 10
EPSS
0.2%
chance of exploitation in 30 days
Higher than 8% of all CVEs
Exploitation Status
Exploit Available
Exploitation: MEDIUM
Sophistication
Moderate
Exploitation Confidence
medium
○CISA SSVC: Public PoC
Composite signal derived from CISA KEV, VulnCheck KEV, CISA SSVC, EPSS, Metasploit, Exploit-DB, trickest/cve, Nuclei templates, and inthewild.io exploitation reports.

What is the attack surface?

AVACPRUISCIA
AVLocal
ACLow
PRLow
UINone
SChanged
CHigh
IHigh
AHigh

What should I do?

1 step
  1. 1) Upgrade MONAI to a release newer than 1.6.0 once a fix is published; monitor GHSA-h89g-r5pc-wxfm and the VulnCheck advisory, since no patched version is currently listed. 2) Until patched, do not run the bundle 'verify_net_in_out' flow on bundles from untrusted or unreviewed sources. 3) Review bundle metadata (shape fields) for non-numeric expressions, attribute access such as '__class__', '__bases__', '__subclasses__', '__globals__' or calls. 4) Run bundle verification in isolated containers or sandboxes with no credentials, least-privilege users and restricted egress. 5) Only pull bundles from trusted, integrity-verified sources (pinned hashes, signatures). 6) Detect via file scanning for dunder patterns in bundle metadata JSON and monitoring for unexpected child processes spawned from MONAI/Python processes.

What does CISA's SSVC say?

DecisionAttend
Exploitationpoc
AutomatableNo
Technical Impacttotal

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:

EU AI Act
Art. 15-Accuracy, robustness and cybersecurity
ISO 42001
A.7.5-Data provenance / supply chain of AI system components
NIST AI RMF
MANAGE 2.3-Responding to unknown risks and managing AI system risks
OWASP LLM Top 10
LLM03-Supply Chain

Frequently Asked Questions

What is CVE-2026-100842?

MONAI through 1.6.0 validates shape expressions in its bundle verification flow with an AST allowlist that only inspects variable names, so expressions built from constants, attributes and calls (for example object-introspection chains) slip through and reach eval(), giving code execution. The flaw is CVSS 8.8 but requires local access with low privileges, and it only triggers in the non-default 'verify_net_in_out' bundle flow when an attacker can influence a bundle's metadata. There is no public exploit, no Nuclei template and it is not in CISA KEV; however, EPSS sits at the 97th percentile (score 0.00148), CISA SSVC says ATTEND, and 126 downstream packages depend on MONAI, so the realistic risk is a poisoned or untrusted bundle (model zoo download, shared pipeline) executing code on a researcher workstation or training host. Treat bundles from untrusted sources as executable code, avoid running 'verify_net_in_out' on them until a fixed release is available (no patched version is listed yet; track GHSA-h89g-r5pc-wxfm), and inventory where MONAI bundles are pulled and verified in your imaging and clinical AI pipelines.

Is CVE-2026-100842 actively exploited?

No confirmed active exploitation of CVE-2026-100842 has been reported, but organizations should still patch proactively.

How to fix CVE-2026-100842?

1) Upgrade MONAI to a release newer than 1.6.0 once a fix is published; monitor GHSA-h89g-r5pc-wxfm and the VulnCheck advisory, since no patched version is currently listed. 2) Until patched, do not run the bundle 'verify_net_in_out' flow on bundles from untrusted or unreviewed sources. 3) Review bundle metadata (shape fields) for non-numeric expressions, attribute access such as '__class__', '__bases__', '__subclasses__', '__globals__' or calls. 4) Run bundle verification in isolated containers or sandboxes with no credentials, least-privilege users and restricted egress. 5) Only pull bundles from trusted, integrity-verified sources (pinned hashes, signatures). 6) Detect via file scanning for dunder patterns in bundle metadata JSON and monitoring for unexpected child processes spawned from MONAI/Python processes.

What systems are affected by CVE-2026-100842?

This vulnerability affects the following AI/ML architecture patterns: training pipelines, model serving, medical imaging AI workflows, model bundle distribution and verification, ML development environments.

What is the CVSS score for CVE-2026-100842?

CVE-2026-100842 has a CVSS v3.1 base score of 8.8 (HIGH). The EPSS exploitation probability is 0.18%.

What is the AI security impact?

Affected AI Architectures

training pipelinesmodel servingmedical imaging AI workflowsmodel bundle distribution and verificationML development environments

MITRE ATLAS Techniques

AML.T0010.003Model
AML.T0011.000Unsafe AI Artifacts
AML.T0018.002Embed Malware
AML.T0050Command and Scripting Interpreter

Compliance Controls Affected

EU AI Act: Art. 15
ISO 42001: A.7.5
NIST AI RMF: MANAGE 2.3
OWASP LLM Top 10: LLM03

What are the technical details?

Original Advisory

MONAI through 1.6.0 contains an eval injection vulnerability in _get_fake_spatial_shape() in monai/bundle/scripts.py. The function validates shape expressions with a helper that walks the AST and only collects ast.Name nodes, rejecting any name other than 'p' or 'n', before passing the string to eval(). Expressions built solely from constants and attribute, subscript, or call nodes (for example "(1).__class__.__bases__[0].__subclasses__()" or "int.__class__.__init__.__globals__") contain no ast.Name nodes and therefore bypass the allowlist. Because the shape value originates from bundle metadata consumed by _get_real_input_data and verify_net_in_out (reachable through the bundle 'verify_net_in_out' CLI flow), an attacker who can influence a bundle's metadata can escape the eval sandbox via object introspection chains and achieve code execution in this non-default flow.

Exploitation Scenario

An attacker publishes or shares a seemingly legitimate MONAI model bundle (for example on a model hub or via a collaborating institution) whose metadata contains a shape expression such as 'int.__class__.__init__.__globals__' chained to reach os or builtins. A data scientist or automated CI job runs the 'verify_net_in_out' flow to validate the bundle; the AST allowlist finds no disallowed names and passes the string to eval(), executing the attacker's payload. The payload steals cloud and registry credentials, accesses training data and model weights on the host, and establishes persistence in the ML environment.

Weaknesses (CWE)

CWE-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:L/UI:N/S:C/C:H/I:H/A:H

Timeline

Published
September 27, 2026
Last Modified
September 30, 2026
First Seen
September 27, 2026

Related Vulnerabilities