Executive brief
Elasticsearch is a widely-used search and analytics platform. A flaw in its machine learning component allows authenticated users with model management privileges to upload specially crafted trained models that execute arbitrary code on the server. This could enable attackers with valid credentials to take complete control of an Elasticsearch cluster, compromising all indexed data and cluster availability.
Technical details
The vulnerability is a deserialization of untrusted data flaw (CWE-502) in Elasticsearch's machine learning component that enables object injection attacks (CAPEC-586). When a specially crafted trained model artifact is uploaded and deployed, attacker-controlled logic executes with elevated system-call privileges. Exploitation requires network access to an Elasticsearch cluster and valid credentials with sufficient permissions to create, upload, and deploy trained models. An authenticated attacker can achieve remote code execution with the privileges of the Elasticsearch process. Mitigation includes disabling machine learning or upgrading to patched versions 8.19.20, 9.4.5, or 9.5.1.
Affected products
- Elastic Elasticsearch 8.0.0 to 8.19.19, 9.0.0 to 9.4.4, 9.5.0
Timeline
- 2026-09-01: disclosed: CVE-2026-72649 published on NVD and ESA-2026-114 advisory released
- 2026-09-01: patched: Fixed in Elasticsearch 8.19.20, 9.4.5, and 9.5.1