Description

Execution with Unnecessary Privileges (CWE-250) in the Kibana machine learning feature can lead to information disclosure via Privilege Abuse (CAPEC-122). An operation available to users holding only read access to the machine learning feature was performed with an internal service identity rather than the identity of the requesting user. Such a user could therefore receive data from Elasticsearch indices they are not authorized to read. No Elasticsearch cluster or index privileges are required.

Executive Summary

Technical Details

Mitigations

References

Additional Metadata

{
  "score": "6.5",
  "pubDate": "2026-09-01T20:17:17.093Z",
  "pubdate": "2026-09-01T20:17:17.093Z",
  "executiveSummary": "This vulnerability involves an Execution with Unnecessary Privileges (CWE-250) flaw within the Kibana machine learning feature, facilitating information disclosure through Privilege Abuse (CAPEC-122).\nThe issue stems from an improper authorization check where operations requested by users with restricted machine learning read-only access are erroneously executed using an internal service identity.\nBy bypassing the requester's actual identity, the system inadvertently grants the user the effective permissions of the privileged service account.\nThis allows unauthorized actors to retrieve data from Elasticsearch indices that are otherwise restricted by their standard security credentials.\nNo specific Elasticsearch cluster or index-level privileges are required for an attacker to exploit this flaw, as the vulnerability resides within the application-level request handling logic of Kibana.\nThe primary risk implication is the potential for unauthorized data exfiltration from sensitive indices, bypassing the principle of least privilege.\nSuccessful exploitation requires minimal prerequisites, as it leverages existing, legitimate access to the machine learning interface, making this a significant threat to data confidentiality within the affected Kibana deployments.",
  "technicalDetails": "The core of this vulnerability lies in the improper delegation of security context when Kibana's machine learning (ML) feature handles user-initiated requests. In a secure architectural pattern, an application must enforce access control lists (ACLs) by verifying the user's identity against the requested resource.\nHowever, in this specific scenario, the machine learning module fails to propagate the authenticated user's session identity to the backend Elasticsearch cluster during certain operations. Instead, the system defaults to an internal service identity—a highly privileged account intended for administrative ML background tasks.\nThe root cause is a failure in the application's request proxying logic, where the identity-switching mechanism does not appropriately constrain the internal service account's scope to the user's intended authorization level. By executing requests under the service identity, the application effectively elevates the user's read capabilities beyond their intended configuration.\nThe attack flow follows a predictable sequence: First, a user with limited 'ML Read' access initiates a specific request through the Kibana ML interface. The server-side component processes this request by invoking an internal service call. Because the call utilizes the internal service identity, it bypasses standard Elasticsearch index-level security controls.\nSince the internal service identity holds elevated rights, Elasticsearch fulfills the data request despite the fact that the original user would normally be forbidden from accessing those indices. The system then returns the sensitive data to the user's browser, effectively completing the information disclosure.\nThe impact is significant because it allows a low-privileged user to query indices containing sensitive, proprietary, or private data without requiring explicit permissions. Because the exploitation relies on the legitimate use of the ML feature, standard anomaly detection or audit logging configured for 'malicious' behavior might fail to flag the request as anomalous, as the request appears to originate from an internal process.\nNo specific network exposure or complex payload is required; the exploitation is performed via standard HTTP(S) requests that invoke the vulnerable Kibana ML functions. This vulnerability exists because the application trust boundaries are misaligned with the intended access control policy, resulting in a systemic bypass of the underlying security infrastructure."
}