CVE-2026-104019: SageMaker Distribution: OS command injection exposes IAM creds
CRITICALA command injection flaw in the Studio Space startup validation script of Amazon SageMaker Distribution (used by SageMaker Unified Studio) lets any authenticated project contributor run arbitrary commands inside another project member's Studio Space and steal that member's temporary execution role credentials. The malicious input is a crafted connection resource property that gets interpolated into a shell call without sanitization. Scored CVSS 9.0 (scope changed, high C/I/A) but exploitation needs low privileges plus user interaction; EPSS is 0.014 (top 28%), it is not in CISA KEV, and there is no public exploit or Nuclei template, so this is not a drop-everything event. The real exposure is lateral movement and privilege escalation across projects that share data and ML roles, so shared Unified Studio domains with many contributors deserve priority. Upgrade to 2.14.12, 3.9.12, 4.0.11, 4.1.11, 4.2.8, 4.3.5 or 4.4.3 depending on your minor line; in Unified Studio, restart Studio Spaces once the patched images are deployed, and migrate off any end-of-support minor lines because they will get no fix.
What is the risk?
Medium-to-high. The technical severity is critical (CVSS 9.0, scope changed) but real-world likelihood is moderate: the attacker must already be an authenticated project contributor and a victim must interact (open or start their Space). EPSS 0.014 (top 28%), no KEV listing, no public exploit and no scanner template lower near-term urgency. Impact is high because the stolen credentials are the victim's execution role, which often has access to S3 datasets, models, and other AWS services. Risk is highest in multi-tenant or multi-team Unified Studio domains with broad contributor access and over-privileged execution roles.
How does the attack unfold?
What systems are affected?
| Package | Ecosystem | Vulnerable Range | Patched |
|---|---|---|---|
| SageMaker | pip | — | No patch |
Do you use SageMaker? You're affected.
How severe is it?
What is the attack surface?
What should I do?
1 step1) Upgrade to the fixed SageMaker Distribution version for your minor line (2.14.12, 3.9.12, 4.0.11, 4.1.11, 4.2.8, 4.3.5, 4.4.3); move off end-of-support minor lines since no patches will be released. 2) In SageMaker Unified Studio, restart Studio Spaces after patched images are deployed so they adopt the latest patch. 3) Review project contributor membership and apply least privilege; remove unneeded contributors. 4) Scope down execution role permissions and enforce permission boundaries/SCPs. 5) Audit connection resources for suspicious property values containing shell metacharacters (;, $(), backticks, |, &&). 6) Hunt in CloudTrail for execution-role credentials used from unexpected IPs or user agents, and for anomalous API calls (S3 listing, STS, IAM enumeration). 7) Rotate/invalidate sessions for potentially affected roles. 8) Follow AWS bulletin 2026-125-AWS and GHSA-w64x-664p-7w66.
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-2026-104019?
A command injection flaw in the Studio Space startup validation script of Amazon SageMaker Distribution (used by SageMaker Unified Studio) lets any authenticated project contributor run arbitrary commands inside another project member's Studio Space and steal that member's temporary execution role credentials. The malicious input is a crafted connection resource property that gets interpolated into a shell call without sanitization. Scored CVSS 9.0 (scope changed, high C/I/A) but exploitation needs low privileges plus user interaction; EPSS is 0.014 (top 28%), it is not in CISA KEV, and there is no public exploit or Nuclei template, so this is not a drop-everything event. The real exposure is lateral movement and privilege escalation across projects that share data and ML roles, so shared Unified Studio domains with many contributors deserve priority. Upgrade to 2.14.12, 3.9.12, 4.0.11, 4.1.11, 4.2.8, 4.3.5 or 4.4.3 depending on your minor line; in Unified Studio, restart Studio Spaces once the patched images are deployed, and migrate off any end-of-support minor lines because they will get no fix.
Is CVE-2026-104019 actively exploited?
No confirmed active exploitation of CVE-2026-104019 has been reported, but organizations should still patch proactively.
How to fix CVE-2026-104019?
1) Upgrade to the fixed SageMaker Distribution version for your minor line (2.14.12, 3.9.12, 4.0.11, 4.1.11, 4.2.8, 4.3.5, 4.4.3); move off end-of-support minor lines since no patches will be released. 2) In SageMaker Unified Studio, restart Studio Spaces after patched images are deployed so they adopt the latest patch. 3) Review project contributor membership and apply least privilege; remove unneeded contributors. 4) Scope down execution role permissions and enforce permission boundaries/SCPs. 5) Audit connection resources for suspicious property values containing shell metacharacters (;, $(), backticks, |, &&). 6) Hunt in CloudTrail for execution-role credentials used from unexpected IPs or user agents, and for anomalous API calls (S3 listing, STS, IAM enumeration). 7) Rotate/invalidate sessions for potentially affected roles. 8) Follow AWS bulletin 2026-125-AWS and GHSA-w64x-664p-7w66.
What systems are affected by CVE-2026-104019?
This vulnerability affects the following AI/ML architecture patterns: ML development environments (managed notebooks/IDE), Multi-tenant ML platforms, Training pipelines, Data lakes and feature stores accessed by ML roles, MLOps platforms.
What is the CVSS score for CVE-2026-104019?
CVE-2026-104019 has a CVSS v3.1 base score of 9.0 (CRITICAL). The EPSS exploitation probability is 1.45%.
What is the AI security impact?
Affected AI Architectures
MITRE ATLAS Techniques
AML.T0012Valid AccountsAML.T0035AI Artifact CollectionAML.T0050Command and Scripting InterpreterAML.T0055Unsecured CredentialsAML.T0106Exploitation for Credential AccessCompliance Controls Affected
What are the technical details?
Original Advisory
OS command injection in the Studio Space startup validation script in Amazon SageMaker Distribution 2.x before 2.14.12, 3.x before 3.9.12, 4.0.x before 4.0.11, 4.1.x before 4.1.11, 4.2.x before 4.2.8, 4.3.x before 4.3.5, and 4.4.x before 4.4.3, as used by Amazon SageMaker Unified Studio, might allow an authenticated remote user with project contributor permissions to execute arbitrary commands in another project member's Studio Space and obtain that member's temporary execution role credentials via a crafted connection resource property that is interpolated into a shell invocation without neutralization. To remediate this issue, users should upgrade to version 2.14.12, 3.9.12, 4.0.11, 4.1.11, 4.2.8, 4.3.5, or 4.4.3, as applicable to the minor line in use. Users on minor lines that have reached end of support must move to a supported minor line, because no patched version will be released for those lines. In Amazon SageMaker Unified Studio, Studio Spaces adopt the latest patch of their minor line on restart once the patched images are deployed, so no version selection is required.
Exploitation Scenario
A contributor on a shared SageMaker Unified Studio project (a malicious insider or an attacker with a compromised contributor account) edits a connection resource, setting a property to a value containing shell metacharacters and a payload. When another project member starts or restarts their Studio Space, the startup validation script interpolates the property into a shell command and runs the payload in the victim's Space. The payload queries the container credential endpoint and exfiltrates the victim's temporary execution role credentials to an attacker-controlled server. The attacker then uses them to read training datasets and model artifacts from S3, modify pipelines, or pivot into other AWS resources the role can reach.
Weaknesses (CWE)
CWE-78Improper Neutralization of Special Elements used in an OS CommandCWE-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:N/AC:L/PR:L/UI:R/S:C/C:H/I:H/A:HReferences
Timeline
Related Vulnerabilities
CVE-2026-1124088.1SageMaker SDK: S3 bucket squatting enables RCE
Same package: sagemakerCVE-2026-835517.2SageMaker: HMAC leak allows forged pipeline code execution
Same package: sagemakerCVE-2026-85977.2SageMaker: RCE via poisoned Triton model artifacts in S3
Same package: sagemakerCVE-2026-17777.2sagemaker: security flaw enables exploitation
Same package: sagemakerCVE-2026-85967.2SageMaker SDK: cleartext HMAC key enables model artifact RCE
Same package: sagemaker