Description

Incorrect Authorization (CWE-863) in Kibana can lead to unauthorized disclosure, modification, and deletion of data via Accessing Functionality Not Properly Constrained by ACLs (CAPEC-1). Where two authenticated principals originating from different authentication realms share the same username value, one could read, modify, and delete the other's private Elastic AI Assistant Knowledge Base entries.

Executive Summary

Technical Details

Mitigations

References

Additional Metadata

{
  "score": "4.2",
  "pubDate": "2026-09-01T20:17:23.887Z",
  "pubdate": "2026-09-01T20:17:23.887Z",
  "executiveSummary": "This vulnerability, classified as CWE-863 (Incorrect Authorization), affects Kibana and relates to how the system handles authentication principals from disparate realms. The issue arises when two distinct authenticated entities share an identical username across different authentication providers, leading to a collision in the access control logic.\nThe vulnerability allows for unauthorized data operations, including the disclosure, modification, and deletion of sensitive information within the Elastic AI Assistant Knowledge Base. An attacker with a valid account in one realm can gain unauthorized access to the private data of another user who shares the same username in a different realm.\nThis represents a significant security risk, as it bypasses intended access control lists (ACLs) through CAPEC-1 (Accessing Functionality Not Properly Constrained by ACLs). The impact includes full read/write/delete capabilities over the victim's private knowledge base entries. Exploitation is limited to authenticated users; however, no administrative privileges are required, provided the naming collision exists between the realms configured for the instance.",
  "technicalDetails": "The root cause of this vulnerability lies in an improper implementation of authorization logic within the Kibana security subsystem. When Kibana evaluates permissions for the Elastic AI Assistant Knowledge Base, it appears to rely primarily on the username attribute to resolve object ownership or access rights, rather than enforcing a unique, realm-qualified identifier.\nIn environments where multiple authentication realms (e.g., LDAP, Active Directory, Native, or SAML) are configured, the system fails to verify the provenance of the authenticated principal correctly when performing authorization checks. If User-A (authenticated via Realm-1) and User-B (authenticated via Realm-2) share the same 'username' string, the underlying access control mechanism conflates the two identities.\nThe attack flow proceeds as follows: First, an attacker authenticates to Kibana using their credentials in their respective realm. Second, the attacker identifies or guesses the existence of a target user in a different realm who shares the same username. Third, the attacker interacts with the Elastic AI Assistant Knowledge Base API. Fourth, the authorization component, failing to distinguish between the two realms due to the reliance on the ambiguous username string, grants the attacker the same access rights as the target user. Consequently, the attacker can execute CRUD (Create, Read, Update, Delete) operations against the target user’s private knowledge base entries.\nThis flaw effectively permits cross-realm privilege escalation or identity spoofing at the application layer. The vulnerable component is the security-context handler responsible for parsing and enforcing ACLs on the Elastic AI Assistant plugin. Since the system does not properly isolate resources based on the combination of the realm and the username, the security boundary is effectively breached for any user sharing a username with a target. The post-exploitation impact is severe, as it facilitates the exfiltration of potentially sensitive documentation or AI context data, as well as the destruction of information stored within the victim's knowledge base."
}