CVE-2026-23536: Jupyter unauth path traversal leaks

CISO Take

Feast Feature Server's /read-document endpoint fails to validate file paths, letting an unauthenticated attacker send a single crafted HTTP POST and read any file the server process can access — configs, credentials, or system files (CWE-22, CVSS 7.5). This isn't a theoretical concern for AI teams: Feast ships as the feature-serving layer in Red Hat OpenShift AI/Open Data Hub, so exposed instances sit inside training pipelines, pipeline-runtime containers, and Jupyter workbenches with 1,889 downstream dependents at risk if the component is internet-facing. The upside is exploitation likelihood is currently modest — EPSS sits at just 0.59% (though that ranks in the top 56% of all scored CVEs), there's no public exploit or Nuclei template, it's not in CISA KEV, and CISA's own SSVC decision is TRACK (lowest urgency tier). Given the trivial exploit mechanics (no auth, no user interaction, low complexity) CISOs should not wait on the low EPSS score: confirm whether Feast Feature Server is reachable from outside your network, check Red Hat's advisory (access.redhat.com/security/cve/CVE-2026-23536) for a patched build, and in the meantime block path-traversal patterns to /read-document at the reverse proxy or WAF and restrict the endpoint to internal traffic only.

Sources: NVD CISA KEV EPSS OpenSSF ATLAS redhat.com

What is the risk?

CVSS 3.1 base score 7.5 (High) reflects a network-exploitable, low-complexity, no-privilege, no-user-interaction vulnerability with high confidentiality impact and no integrity/availability impact (AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:N) — a pure information-disclosure bug, not RCE. Real-world exploitation pressure is currently low: EPSS is 0.594% (top 56th percentile, not a high absolute probability), there is no public PoC or Nuclei template, the CVE is absent from CISA KEV, and CISA's SSVC decision is TRACK, its lowest action tier. However, ease of exploitation (a single unauthenticated POST) combined with the package's OpenSSF Scorecard of 5.8/10 and history of 26 other CVEs in the same package suggests a codebase with recurring hardening gaps. Exposure risk depends entirely on network placement — feature servers deployed behind internal networks or service meshes have materially lower real-world risk than internet-facing deployments, which is the key variable CISOs need to verify.

How does the attack unfold?

Discovery

Attacker scans for internet-facing ML infrastructure and identifies an exposed Feast Feature Server /read-document endpoint.

AML.T0006

Exploitation

A crafted HTTP POST with a path-traversal payload bypasses intended access restrictions and returns arbitrary file contents from the server.

AML.T0049

Credential Harvesting

Attacker reads configuration files or secrets from the local filesystem exposed via the traversal bug.

AML.T0055

Pivot / Impact

Harvested credentials are used to authenticate to connected databases or cloud services, extending the compromise beyond the feature server.

AML.T0012

Discovery

Attacker scans for internet-facing ML infrastructure and identifies an exposed Feast Feature Server /read-document endpoint.

AML.T0006

Exploitation

A crafted HTTP POST with a path-traversal payload bypasses intended access restrictions and returns arbitrary file contents from the server.

AML.T0049

Credential Harvesting

Attacker reads configuration files or secrets from the local filesystem exposed via the traversal bug.

AML.T0055

Pivot / Impact

Harvested credentials are used to authenticate to connected databases or cloud services, extending the compromise beyond the feature server.

AML.T0012

What systems are affected?

Package	Ecosystem	Vulnerable Range	Patched
Jupyter	pip	—	No patch
13.3K OpenSSF 5.7 1.9K dependents Pushed 6d ago 54% patched ~28d to patch Full package profile →
Jupyter	pip	—	No patch
13.3K OpenSSF 5.7 1.9K dependents Pushed 6d ago 54% patched ~28d to patch Full package profile →
Jupyter	pip	—	No patch
13.3K OpenSSF 5.7 1.9K dependents Pushed 6d ago 54% patched ~28d to patch Full package profile →
Jupyter	pip	—	No patch
13.3K OpenSSF 5.7 1.9K dependents Pushed 6d ago 54% patched ~28d to patch Full package profile →
Jupyter	pip	—	No patch
13.3K OpenSSF 5.7 1.9K dependents Pushed 6d ago 54% patched ~28d to patch Full package profile →
Jupyter	pip	—	No patch
13.3K OpenSSF 5.7 1.9K dependents Pushed 6d ago 54% patched ~28d to patch Full package profile →
rhoai/odh-feature-server-rhel9	—	—	No patch
rhoai/odh-pipeline-runtime-datascience-cpu-py312-rhel9	—	—	No patch
rhoai/odh-pipeline-runtime-pytorch-cuda-py312-rhel9	—	—	No patch
rhoai/odh-pipeline-runtime-pytorch-llmcompressor-cuda-py312-rhel9	—	—	No patch
rhoai/odh-pipeline-runtime-pytorch-rocm-py312-rhel9	—	—	No patch
rhoai/odh-pipeline-runtime-tensorflow-cuda-py312-rhel9	—	—	No patch
rhoai/odh-pipeline-runtime-tensorflow-rocm-py312-rhel9	—	—	No patch
rhoai/odh-workbench-codeserver-datascience-cpu-py312-rhel9	—	—	No patch

How severe is it?

CVSS 3.1

7.5 / 10

EPSS

0.6%

chance of exploitation in 30 days

Higher than 46% of all CVEs

Source: EPSS v3 — FIRST.org

Exploitation Status

No known exploitation

Sophistication

Trivial

What is the attack surface?

AV Network

AC Low

PR None

UI None

S Unchanged

C High

I None

A None

What should I do?

1 step

1) Check Red Hat's advisory (access.redhat.com/security/cve/CVE-2026-23536) and Bugzilla #2429302 for a patched RHOAI/ODH image build and upgrade affected feature-server and pipeline-runtime containers as soon as one is available — no patched version is listed in current data, so track vendor release notes closely. 2) Until patched, do not expose Feast Feature Server directly to the internet; place it behind a network boundary, service mesh, or authenticating reverse proxy so /read-document is unreachable from untrusted networks. 3) Add WAF/reverse-proxy rules blocking path-traversal sequences (../, URL-encoded variants, null bytes) in POST bodies to /read-document. 4) Run the feature server process with a minimal-privilege OS user and container filesystem restrictions (read-only root FS, no mounted secrets beyond what's required) to cap the blast radius of any successful file read. 5) Monitor feature-server access logs for anomalous file paths, repeated traversal attempts, or requests targeting /etc/passwd, .env, or cloud metadata IPs (169.254.169.254) in the /read-document payload. 6) Rotate any credentials stored in configuration files reachable by the server process, since exposure cannot be ruled out retroactively.

What does CISA's SSVC say?

Decision Track

Exploitation none

Automatable Yes

Technical Impact partial

Source: CISA Vulnrichment (SSVC v2.0). Decision based on the CISA Coordinator decision tree.

How is it classified?

Data Extraction Auth Bypass RAG Inference Framework AML.T0037 - Data from Local System AML.T0049 - Exploit Public-Facing Application AML.T0055 - Unsecured Credentials

Which compliance frameworks are affected?

This CVE is relevant to:

EU AI Act

Article 15 - Accuracy, robustness and cybersecurity

ISO 42001

Annex A.6 - Security controls for AI system components

NIST AI RMF

MEASURE 2.7 - AI system security and resilience evaluated and documented

OWASP LLM Top 10

LLM02 - Sensitive Information Disclosure

Frequently Asked Questions

What is CVE-2026-23536?

Feast Feature Server's /read-document endpoint fails to validate file paths, letting an unauthenticated attacker send a single crafted HTTP POST and read any file the server process can access — configs, credentials, or system files (CWE-22, CVSS 7.5). This isn't a theoretical concern for AI teams: Feast ships as the feature-serving layer in Red Hat OpenShift AI/Open Data Hub, so exposed instances sit inside training pipelines, pipeline-runtime containers, and Jupyter workbenches with 1,889 downstream dependents at risk if the component is internet-facing. The upside is exploitation likelihood is currently modest — EPSS sits at just 0.59% (though that ranks in the top 56% of all scored CVEs), there's no public exploit or Nuclei template, it's not in CISA KEV, and CISA's own SSVC decision is TRACK (lowest urgency tier). Given the trivial exploit mechanics (no auth, no user interaction, low complexity) CISOs should not wait on the low EPSS score: confirm whether Feast Feature Server is reachable from outside your network, check Red Hat's advisory (access.redhat.com/security/cve/CVE-2026-23536) for a patched build, and in the meantime block path-traversal patterns to /read-document at the reverse proxy or WAF and restrict the endpoint to internal traffic only.

Is CVE-2026-23536 actively exploited?

No confirmed active exploitation of CVE-2026-23536 has been reported, but organizations should still patch proactively.

How to fix CVE-2026-23536?

1) Check Red Hat's advisory (access.redhat.com/security/cve/CVE-2026-23536) and Bugzilla #2429302 for a patched RHOAI/ODH image build and upgrade affected feature-server and pipeline-runtime containers as soon as one is available — no patched version is listed in current data, so track vendor release notes closely. 2) Until patched, do not expose Feast Feature Server directly to the internet; place it behind a network boundary, service mesh, or authenticating reverse proxy so /read-document is unreachable from untrusted networks. 3) Add WAF/reverse-proxy rules blocking path-traversal sequences (../, URL-encoded variants, null bytes) in POST bodies to /read-document. 4) Run the feature server process with a minimal-privilege OS user and container filesystem restrictions (read-only root FS, no mounted secrets beyond what's required) to cap the blast radius of any successful file read. 5) Monitor feature-server access logs for anomalous file paths, repeated traversal attempts, or requests targeting /etc/passwd, .env, or cloud metadata IPs (169.254.169.254) in the /read-document payload. 6) Rotate any credentials stored in configuration files reachable by the server process, since exposure cannot be ruled out retroactively.

What systems are affected by CVE-2026-23536?

This vulnerability affects the following AI/ML architecture patterns: RAG pipelines, feature stores, model serving, training pipelines.

What is the CVSS score for CVE-2026-23536?

CVE-2026-23536 has a CVSS v3.1 base score of 7.5 (HIGH). The EPSS exploitation probability is 0.59%.

What is the AI security impact?

Affected AI Architectures

RAG pipelinesfeature storesmodel servingtraining pipelines

MITRE ATLAS Techniques

AML.T0037 Data from Local System

AML.T0049 Exploit Public-Facing Application

AML.T0055 Unsecured Credentials

Compliance Controls Affected

EU AI Act: Article 15

ISO 42001: Annex A.6

NIST AI RMF: MEASURE 2.7

OWASP LLM Top 10: LLM02

What are the technical details?

Original Advisory

A security issue was discovered in the Feast Feature Server's `/read-document` endpoint that allows an unauthenticated remote attacker to read any file accessible to the server process. By sending a specially crafted HTTP POST request, an attacker can bypass intended access restrictions to potentially retrieve sensitive system files, application configurations, and credentials.

Exploitation Scenario

An attacker scanning internet-facing ML infrastructure discovers a Red Hat OpenShift AI deployment exposing a Feast Feature Server. They identify the /read-document endpoint — intended to serve indexed documents to a RAG pipeline — and send a crafted HTTP POST with a path-traversal payload (e.g., referencing ../../../../etc/passwd or an internal .env/config path) instead of a legitimate document ID. Because the endpoint doesn't validate or restrict the requested path (CWE-22), the server returns the raw file contents with no authentication required. The attacker harvests database credentials or cloud service-account tokens from configuration files exposed this way, then uses those credentials to authenticate directly to the organization's data stores or cloud accounts, pivoting well beyond the feature server itself.

Weaknesses (CWE)

CWE-22 Improper Limitation of a Pathname to a Restricted Directory ('Path Traversal')

CWE-22 — Improper Limitation of a Pathname to a Restricted Directory ('Path Traversal'): The product uses external input to construct a pathname that is intended to identify a file or directory that is located underneath a restricted parent directory, but the product does not properly neutralize special elements within the pathname that can cause the pathname to resolve to a location that is outside of the restricted directory.

[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
[Architecture and Design] For any security checks that are performed on the client side, ensure that these checks are duplicated on the server side, in order to avoid CWE-602. Attackers can bypass the client-side checks by modifying values after the checks have been performed, or by changing the client to remove the client-side checks entirely. Then, these modified values would be submitted to the server.

Source: MITRE CWE corpus.