CVE-2026-72675

This CVE exposes a fundamental design weakness in Kibana's space isolation that goes well beyond a single missing code filter. The ML functionality failed to apply the space context filter to its Elasticsearch queries — a filter that every other code path in Kibana must manually remember to apply. Because ML operations run with elevated internal permissions, this single omission allowed cross-space data access across every space in the deployment. The vulnerability isn't that one developer forgot a line of code. It's that Kibana's security model relies on developer discipline rather than architectural enforcement. The space boundary exists as a per-request filter that each code path must consciously apply — an implicit security model that depends on every team, every refactor, and every new engineer remembering the convention. This is the same pattern that produced decades of SQL injection and access control failures: a convention that works perfectly on the happy path, fails invisibly on the unusual case, and produces severe consequences when it breaks. The critical architectural question is whether ML's elevated internal permissions were necessary for the feature to function, or whether they were granted as institutional shortcut that created an amplification ladder for authorization failures. The fix (adding the space filter to the ML code path) addresses the symptom, not the structural disease. Organizations running Kibana should treat this CVE as evidence that convention-based security boundaries in Kibana may have similar latent failures in other code paths — particularly in feature areas that were added after spaces were established as the isolation model. Practical steps: verify your Kibana version is patched, audit other ML code paths for consistent space filtering, and evaluate whether the organizational process for ML feature development includes explicit security review of authorization assumptions. The EPSS score of 0.00215 suggests moderate exploitation likelihood, but the vulnerability class (broken authorization via privilege abuse) historically sees higher exploitation rates than initially predicted.

Reviewed through automated stages and approved by a human before publication.