Kibana CVE-2026-72671

Exploitation requires an authenticated Kibana session for a user whose RBAC role grants the create anomaly detection jobs and data frame analytics jobs privileges but does not grant the trained model privilege. The target space must contain at least one trained model saved object. The flaw is scoped exclusively to the ML remove-saved-object-from-space operation and does not affect other Kibana features or object types. Exploitation is limited by: (1) mandatory authenticated session with a specifically scoped ML role - unauthenticated or unprivileged users cannot exploit this; (2) the trained model is not deleted and remains accessible in other spaces; (3) the action is fully reversible by a user with trained model management privileges. Default Kibana deployments with no custom ML roles configured or with standard superuser roles are not meaningfully exposed, as superusers already hold all privileges including trained model access.

The vendor-assigned CVSS 3.1 score of 4.3 (Medium) with vector AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:L/A:N accurately reflects the limited blast radius: network-accessible, low complexity, but requires authenticated access (PR:L) and produces only low integrity impact with no confidentiality or availability loss. EPSS data is absent from available intelligence and cannot be incorporated into this assessment. The vulnerability is not listed in CISA KEV and no POC is referenced, placing it at SSVC exploitation status 'none' at time of analysis. The practical risk is further constrained by the reversible nature of the impact - the trained model is not deleted and a sufficiently privileged user can restore its space membership - and by the prerequisite of a deliberately scoped RBAC configuration granting ML job creation without trained model access. This is a real authorization bypass but not a critical-priority issue for most environments; organizations with strict ML governance or compliance requirements around model promotion workflows should treat it with elevated urgency.

An authenticated Kibana user with a role granting ML anomaly detection and data frame analytics job creation privileges - but lacking explicit trained model access - navigates to the Machine Learning space management interface and invokes the remove-from-space action targeting a deployed trained model. The privilege check passes due to the incomplete authorization logic, and the model is removed from the active space, potentially breaking inference pipelines, dashboards, or alerting rules that reference it. A privileged administrator can restore the model's space membership, but the disruption persists until discovered and remediated.