CVE-2026-63145
MediumCVSS 4.3Exploitation Probability (EPSS)
Low risk6th percentile - higher than 6% of all known CVEs
Summary
An incorrect authorization vulnerability in Kibana's Machine Learning functionality allows low-privileged users to manipulate ML audit and notification records for arbitrary jobs across spaces using internally elevated credentials.
Risk Assessment
An attacker could compromise the integrity of ML records, potentially hiding unauthorized actions or manipulating machine learning outcomes.
Recommendation
Apply the Kibana update that fixes authorization checks for ML management endpoints.
Other vulnerabilities in Kibana
See all- CVE-2017-8452High
Kibana versions prior to 5.2.1 configured for SSL client access do not clean up file descriptors after certain requests, leading to accumulation over time until the process crashes.
- CVE-2016-1000219High
Kibana versions before 4.5.4 and 4.1.11, when a custom output is configured for logging, may write cookies and authorization headers to log files. This information could be used to hijack sessions of other users using Kibana behind some form of authentication such as Shield.
- CVE-2026-78581Medium
Authorization Bypass Through User-Controlled Key (CWE-639) in Kibana can lead to unauthorized data modification via Accessing Functionality Not Properly Constrained by ACLs. Under certain conditions, an authenticated user could reference another user's AI Assistant conversation identifier to access or modify a conversation they do not own. Successful exploitation requires knowledge of a hard-to-guess identifier.
- CVE-2026-72681Medium
Kibana Agent Builder does not correctly verify that the requesting user holds the privileges required by a separate Kibana feature before it creates and runs a tool that invokes that feature's functionality. This allows privilege escalation and could lead to disclosure of sensitive information that the user is not authorized to read.
- CVE-2026-72680Medium
Kibana Agent Builder A2A JSON-RPC API endpoint derives the identifier of a stored conversation from a user-supplied input, and the ownership check on that identifier does not distinguish between a conversation that does not exist and one that exists but belongs to another user. As a result, an authenticated user holding only the Agent Builder read privilege can supply an identifier already in use by another user in the same space and cause that user's conversation to be replaced and reassigned to the requesting account. The original owner permanently loses access to the conversation and its history. The impact is limited to loss of integrity and availability of the affected conversation; the attacker does not read the overwritten content.
- CVE-2026-72677High
Kibana contains a Relative Path Traversal (CWE-23) vulnerability that can lead to unauthorized deletion of Kibana resources. Kibana Fleet accepted a user-supplied identifier for a Fleet Server host configuration without rejecting relative traversal sequences. The identifier is stored and later incorporated into the request when that configuration is removed.
- CVE-2026-72675High
Kibana contains a Missing Authorization (CWE-862) vulnerability that can lead to cross-space information disclosure and unauthorized data modification. Kibana Machine Learning carries out its Elasticsearch operations with elevated internal permissions and relies on a per-request space filter to keep ML data separated. Part of the ML functionality did not apply that filter, so operations from one space were carried out against ML data of every space.
- CVE-2026-72674Medium
Kibana has a resource allocation vulnerability (CWE-770) leading to denial of service via excessive allocation. A user-supplied list of document fields in the Playground for RAG feature is not bounded or de-duplicated before building responses. A single crafted request can make Kibana build a response far larger than the source data, exhausting instance resources.
- CVE-2026-72673Medium
Kibana has an incorrect authorization vulnerability (CWE-863) allowing unauthorized deletion of Synthetics private locations. Locations can be shared across spaces, and the deletion safeguard only checks monitors visible in the user's own space. An authenticated user with Synthetics write privilege in one space can delete a location that other spaces depend on, even without access to those spaces.
- CVE-2026-72672High
The Elastic Security capability that suggests existing field values while a user authors endpoint policy artifacts queries Elastic Defend event data with Kibana's internal Elasticsearch account instead of the requesting user's account. Only Kibana feature privileges are verified, not the caller's Elasticsearch index privileges. An authenticated user with Elastic Security feature privileges but no read access to Elastic Defend event indices can retrieve field values, including process command line arguments, which commonly contain tokens, credentials, and other sensitive operational detail.
Original NVD description (English source)
Incorrect Authorization (CWE-863) in Kibana can lead to integrity compromise of Machine Learning audit and notification records via Accessing Functionality Not Properly Constrained by ACLs (CAPEC-1). A vulnerability exists in Kibana's Machine Learning functionality where a Machine Learning management endpoint performs an insufficient authorization check. The endpoint validates only a coarse privilege level but does not verify that the requesting user has access to the specific Machine Learning job or notification resources provided in the request. As a result, a low-privileged user with Machine Learning access in any Kibana space can manipulate Machine Learning audit and notification records for arbitrary jobs—including jobs in other spaces or belonging to other users—by leveraging Kibana's internally elevated credentials to write to restricted Machine Learning system indices that the user cannot access directly.