Server-Side Request Forgery (CVE-2026-64849) in MLflow Webhook Engine Enables Authenticated Data Exposure
What It Is — MLflow (v ≥ 2.10.0) contains a server‑side request forgery (SSRF) flaw in its webhook test endpoint. An authenticated attacker can supply an unchecked URI, causing the MLflow service account to issue arbitrary outbound HTTP requests. The vulnerability can be used to probe internal services, exfiltrate data reachable from the host, or pivot to other internal resources. CVSS 6.8 (moderate) with network‑vector, low‑complexity, required authentication, and high confidentiality impact. MLflow has published a patch (see GitHub PR #24258).
Affected Products — MLflow (all versions prior to the security‑release referenced in the advisory).
Why It Matters for Trust & Control Assurance — The issue underscores the need for rigorous input‑validation controls and continuous monitoring of service‑account activity—an objective that maps to multiple frameworks (e.g., NIST CSF Protect, ISO 27001 A.12).
Recommended Actions —
- Apply the MLflow update that fixes CVE‑2026‑64849 immediately.
- Restrict webhook test functionality to trusted IP ranges or disable it if not required.
- Enforce least‑privilege for the MLflow service account and audit its outbound traffic.
- Add logging and alerting for unusual outbound requests from the MLflow host.
- Incorporate the remediation into your secure‑development and configuration‑management controls.