CVE-2026-64849
mlflow mlflow
Summary
MLflow is an open source AI engineering platform for agents, large language models, and machine learning models. Starting in 3.3.0 and prior to 3.15.0, the unauthenticated POST /api/2.0/mlflow/webhooks/{id}/test endpoint calls _validate_we…
Published: 2026-08-17 · In KEV since: 2026-08-19
Analysis
🔴 Critical — actively exploited
Weakness: Server-Side Request Forgery (SSRF) (CWE-918). Forces the server to make requests to internal targets.
How it's exploited: Exploitable over the network (internet-reachable), unauthenticated, no user interaction, and with low complexity. The flaw can pivot to other components (scope changed).
⚠️ Critical combo: remote, unauthenticated and no interaction → trivial to automate.
Impact: Confidentiality high · Integrity partial · Availability none.
Status: Active exploitation confirmed by CISA (KEV).
- Patch now: it is in CISA's actively-exploited catalog (KEV), with a deadline.
- Reduce exposure: keep it off the internet; firewall, VPN or network segmentation.
- Detect: turn this CVE into detection rules.
MITRE ATT&CK technique: T1190 · Exploit Public-Facing Application
Practice it in the arsenal: WAF Bypass
CVSS vector
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:C/C:H/I:L/A:N
The vector encodes how it's exploited (translated above, in the analysis). Break it down in the CVSS tool →
Sources: CISA KEV · EPSS · FIRST.org · NVD. Analysis derived from the CVSS vector and exploitation status (no AI). Informational; always verify against the vendor's official advisory.