Attack ChainAIDerived
Hypothetical attack flow derived from CVE metadata – not a confirmed exploit chain
-
AccessReach frontend /_proxy/ endpoint unauthenticated
-
DeliveryCraft request or Referer to internal target
-
ExploitServer proxies to 169.254.169.254 or K8s API
-
ExecutionForward attacker headers and body upstream
-
PersistReturn metadata credentials or API data
-
ImpactPivot to read/modify cluster resources
Vulnerability Assessment
| Weakness |
CWE
OWASP Top 10:2025
A01:2025 - Broken Access Control
ASVS 5.0
V8 - Authorization Access control enforcement
MITRE ATT&CK
1 technique across 1 tactic: Defense Evasion The technique list with ATT&CK links and the kill-chain view is part of the Basic plan and up. |
| Exploitation Conditions | Exploitation requires only network reachability to the Kubeflow Pipelines frontend server; the /_proxy/ route is unauthenticated and remains outside the authorization middleware even when ENABLE_AUTHZ=true, so enabling authorization does NOT mitigate it. The attacker controls the target via the request path or a crafted Referer header on any of /apis/v1beta1/_proxy/, /apis/v2beta1/_proxy/, /pipeline/apis/v1beta1/_proxy/, or /pipeline/apis/v2beta1/_proxy/. Limiting factors: the attacker must be able to reach the frontend server's network endpoint, and the practical impact depends on what internal services are reachable from the frontend pod (e.g. an accessible cloud metadata endpoint or Kubernetes API); egress network restrictions or IMDSv2 on the pod materially constrain what can be exfiltrated. Affected versions are all releases prior to 2.17.0. |
| Impact | All available signals point to genuine critical risk rather than an inflated CVSS. The vector CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:C/C:H/I:H/A:H is internally consistent with the description: network-reachable, low complexity, no privileges, no user interaction, and a scope change because the frontend server is abused to pivot into a different security authority (internal cluster/cloud services). The confidentiality and integrity impacts are well supported - the proxy forwards attacker-chosen methods and Authorization/Cookie headers and returns upstream responses, enabling both reading (metadata credentials) and modifying (service/K8s API calls) internal resources; availability impact is plausible via destructive API calls though slightly less direct than C/I. No EPSS score, KEV entry, or public PoC was provided, so exploitation is not confirmed in the wild; however the vulnerability class (unauthenticated SSRF to cloud metadata) is a well-understood, commonly weaponized pattern, so absence of a published PoC should not lower urgency. Net: treat as a top-priority patch, especially for internet-exposed or multi-tenant clusters. |
| Exploit ScenarioDerived | An attacker who can reach the Kubeflow Pipelines UI sends an unauthenticated request to /apis/v1beta1/_proxy/ (or supplies a crafted Referer header) pointing at http://169.254.169.254/latest/meta-data/iam/security-credentials/, and the frontend server proxies the request and returns cloud IAM credentials in the response. The same primitive can be aimed at the in-cluster Kubernetes API or other internal services with attacker-chosen methods and headers to read or modify resources. No PoC is published, but the AV:N/AC:L/PR:N/UI:N profile means exploitation requires only crafted HTTP requests and no authentication. |
Get an e-mail when this CVE is exploited, patched or re-scored. Free account.
Alert meAnalysisAIv1 – Aug 28, 2026
Server-side request forgery in the Kubeflow Pipelines frontend server (prior to 2.17.0) lets any unauthenticated network client coerce the pipeline UI backend into making arbitrary HTTP/HTTPS requests to internal targets via the /_proxy/ route. Because the route sits outside the authorization middleware even when ENABLE_AUTHZ=true, attackers can reach cloud metadata endpoints, the Kubernetes API, and other cluster-internal services, forwarding chosen methods, Authorization/Cookie headers, and POST bodies while receiving the upstream response. No public exploit identified at time of analysis, but the CVSS 10.0 rating and default reachability make this a critical priority for any exposed Kubeflow deployment.
RemediationAI
Upgrade to Kubeflow Pipelines 2.17.0, which removes the generic /_proxy/ endpoint (it now returns HTTP 410) and replaces it with a scoped, secret-signed TensorBoard proxy (Vendor-released patch: 2.17.0). Apply via the fix in PR https://github.com/kubeflow/pipelines/pull/13511 / commit a35f97aa4c17b25572369e5022546ab4421bdbdd, per advisory GHSA-gqww-5pj5-8fq7. If immediate upgrade is not possible, restrict access to the frontend server at the network layer - block or deny the /_proxy/ paths (/apis/v1beta1/_proxy/, /apis/v2beta1/_proxy/, /pipeline/apis/v1beta1/_proxy/, /pipeline/apis/v2beta1/_proxy/) at an ingress/reverse proxy or NetworkPolicy, noting that legacy UI features relying on the proxy (e.g. TensorBoard viewing) may break. Additionally, enforce egress NetworkPolicies on the frontend pod to deny traffic to the cloud metadata IP (169.254.169.254), loopback, and the Kubernetes API server, and where supported enable IMDSv2 / hop-limit restrictions so metadata credentials cannot be retrieved via SSRF; these controls reduce but do not fully eliminate exposure to other in-cluster services.
Affected VersionsENISA
Affected Packages
Source: OSV (Open Source Vulnerabilities)
Affects your stack
Affects 1 package in euvd
Checking this CVE against your own stacks and servers is part of the Basic plan and up.
Technical ContextAI · show details
Kubeflow Pipelines is a component of the Kubeflow ML platform that runs an Express/Node.js frontend server (frontend/server/proxy-middleware.ts) on Kubernetes. The flaw lives in the _routePathWithReferer() helper, which takes an attacker-controlled path or Referer header, prepends http:// if no scheme is present, and passes the resulting URL origin straight to http-proxy-middleware's createProxyMiddleware. There is no host allowlist and no filtering of loopback (127.0.0.0/8), link-local (169.254.0.0/16, the cloud metadata range), RFC1918 private, or cluster-local addresses. CWE-284 (Improper Access Control) is the root cause: the /_proxy/ endpoints - reachable at /apis/v1beta1/_proxy/, /apis/v2beta1/_proxy/, and their /pipeline/ prefixed variants - are registered outside the authorization layer, so the access-control gate that protects the rest of the API never applies. The fix (PR #13511) removes the generic proxy entirely, returning HTTP 410 and replacing it with a scoped, secret-signed TensorBoard-only proxy.
Know when Kubeflow gets its next vulnerability
- Watch products and get an email when something new is published
- Check this CVE against your own stack and servers
- Exploit and PoC references, threat intelligence from 22 sources
No credit card · 7-day full trial
Severity by source
AV:N/AC:L/PR:N/UI:N/S:C/C:H/I:H/A:H
Unauthenticated network-reachable SSRF (AV:N/AC:L/PR:N/UI:N) pivots the frontend into a different authority (S:C), enabling reads of credentials and reads/writes to internal APIs (C:H/I:H/A:H).
Primary rating from Vendor (GitHub_M).
CVSS VectorVendor: GitHub_M
Lifecycle Timeline
2DescriptionCVE.org
Kubeflow Pipelines enables users to build and deploy portable, scalable machine learning workflows. Prior to 2.17.0, the Kubeflow Pipelines frontend exposes an unauthenticated server-side request forgery vulnerability through the /_proxy/ route in frontend/server/proxy-middleware.ts. The _routePathWithReferer() function accepts an arbitrary attacker-controlled HTTP or HTTPS target and passes its origin to createProxyMiddleware without a host allowlist or filtering for loopback, link-local, RFC1918, or cluster-local addresses. The route remains outside the authorization middleware when ENABLE_AUTHZ=true and is reachable through /apis/v1beta1/_proxy/, /apis/v2beta1/_proxy/, /pipeline/apis/v1beta1/_proxy/, and /pipeline/apis/v2beta1/_proxy/, including through a crafted Referer header. Requests can forward attacker-controlled methods, headers such as Authorization, Cookie, and X-Forwarded-For, and POST bodies to reachable internal services, while returning the upstream response to the unauthenticated client. This can expose cloud metadata credentials, Kubernetes or service APIs, and other cluster-internal endpoints to unauthorized read or modification. This issue is fixed in version 2.17.0.
Articles & Coverage 2
Technical POC Analysis
Server-side request forgery, CWE-918, with an authorization gap, CWE-284.
Full technical POC analysis - root cause, step-by-step exploitation and patch breakdown - is part of the Pro and Business plans.
Executive ViewAI
A critical server-side request forgery in Kubeflow Pipelines frontend (CVE-2026-54745, CVSS 10.0) allows unauthenticated attackers to bypass authentication controls and make arbitrary requests to internal Kubernetes APIs and cloud metadata services. Any organization running this ML infrastructure component faces immediate risk of cluster compromise, credential exposure, and lateral movement into cloud infrastructure. No vendor patch is currently available, requiring immediate interim network-level mitigation.
Block network access to the /_proxy/ endpoint using firewall rules or reverse proxy, implement Web Application Firewall (WAF) rules to reject requests to /_proxy/*, restrict Kubeflow frontend network access to authorized users and systems only, configure pod-level egress network policies to limit internal service communication, and monitor for CISA KEV listing and exploitation indicators.
Threat Intelligence Overview
How this decision was reached
The engine reads each signal below and applies a fixed list of rules from top to bottom. The first rule that matches sets the recommended action – like a statement, every line is shown.
What every action level means
SSVC AssessmentENISA
Priority Score
Attack Technique
Server-Side Request Forgery exploits applications that fetch remote resources based on user-supplied URLs.
Authentication bypass attacks exploit flaws in the verification mechanisms that control access to systems and applications.
kubeflow/kubeflow is vulnerable to a Regular Expression Denial of Service (ReDoS) attack due to inefficient regular expr
Server-Side Request Forgery (SSRF) in kubeflow/kubeflow. Rated medium severity (CVSS 6.5), this vulnerability is remotel
Cross-site Scripting (XSS) - Reflected in kubeflow/kubeflow. Rated medium severity (CVSS 6.1), this vulnerability is rem
Authorization token theft in Kubeflow Community Distribution before 26.03-rc.1 lets a low-privileged tenant hijack other
Kubeflow Model Registry through 0.3.17 allows an authenticated namespace user with permission to create InferenceService
There exists a stored XSS Vulnerability in Kubeflow Pipeline View web UI. Rated high severity (CVSS 7.1), this vulnerabi
EUVD-2026-67790
NVD
CVE.org
EPSS
Vulnrichment
GitHub
Exploit-DB