Description

Authorization Bypass Through User-Controlled Key (CWE-639) in Kibana can lead to unauthorized cross-space access via Accessing Functionality Not Properly Constrained by ACLs (CAPEC-1). The result is disclosure of inference output from a trained model in a different space that the user is not authorized to list, read, or use, which exposes the behavior of a model. The same pattern also reached the deployment stop and deployment update operations, allowing an active trained model deployment in another space to be stopped or to have its allocated resources altered.

Executive Summary

Technical Details

Mitigations

References

Additional Metadata

{
  "score": "7.1",
  "pubDate": "2026-08-13T20:17:23.690Z",
  "pubdate": "2026-08-13T20:17:23.690Z",
  "executiveSummary": "An Authorization Bypass Through User-Controlled Key vulnerability, classified as CWE-639, has been identified in Kibana.\nThis vulnerability maps to CAPEC-1 (Accessing Functionality Not Properly Constrained by ACLs) and enables unauthorized cross-space access to system functionality.\nThe primary impact of this flaw includes the unauthorized disclosure of inference output derived from trained models residing in separate spaces that the attacking user lacks authorization to list, read, or use, thereby exposing proprietary model behavior.\nFurthermore, the vulnerability extends to deployment management operations, specifically deployment stop and deployment update functions.\nConsequently, an authenticated attacker with access to manipulated user-controlled keys can cause an active trained model deployment in an unauthorized space to be terminated or have its allocated resources improperly altered.\nThe risk implications involve a compromise of data confidentiality and operational integrity across isolated Kibana spaces, potentially leading to service disruption and intellectual property exposure regarding machine learning models.\nExploitation requires the capability to manipulate user-controlled keys within requests to interact with spaces and functionalities beyond the attacker's assigned access control lists.",
  "technicalDetails": "The root cause of the vulnerability stems from insufficient authorization checks and improper access control enforcement on user-controlled keys within Kibana spaces.\nWhen interacting with machine learning model functionalities, the underlying application logic fails to properly validate whether the requesting user possesses the necessary permissions via Access Control Lists (ACLs) to access specific spaces containing trained models and deployments.\nThe vulnerability manifests through inadequate server-side validation of object identifiers or keys supplied by the user during API requests.\nThe attack flow proceeds as follows: an authenticated user crafts an HTTP request targeting a model inference endpoint, substituting or manipulating the user-controlled key to reference an unauthorized target space.\nBecause the application relies on the user-supplied key without adequately verifying boundaries enforced by ACLs, it processes the request within the context of the unauthorized space.\nThis flaw allows the attacker to retrieve sensitive inference output, exposing the behavior of a trained model residing in a space they are not authorized to list, read, or use.\nAdditionally, the same architectural flaw affects deployment lifecycle management endpoints, specifically deployment stop and deployment update operations.\nBy leveraging the authorization bypass pattern against these endpoints, an active trained model deployment located in another space can be maliciously stopped or subjected to unauthorized resource alterations.\nThe vulnerable components involve the access control and routing mechanisms handling multi-tenancy spaces and machine learning deployment functions within Kibana.\nExploitation requires network exposure to the Kibana instance, valid authentication credentials to interact with the application, and the ability to craft requests containing manipulated user-controlled keys to bypass intended isolation boundaries."
}