Citrix and MLflow Flaws Face Active Exploitation

MLflow, an open-source platform for tracking experiments and managing machine-learning models through deployment, is often connected to sensitive corporate cloud infrastructure. The server-side request forgery flaw, designated CVE-2026-64849, carries a CVSS severity score of 9.3. Successful exploitation could allow an unauthenticated attacker to make requests through a vulnerable MLflow server, probe internal services and reach cloud metadata endpoints that may expose credentials or other confidential information.

Security firm watchTowr said it detected large-scale scanning within hours of the vulnerability’s public disclosure, signaling that attackers were rapidly searching for internet-exposed MLflow instances. U.S. cybersecurity authorities later warned that the flaw was being actively exploited. The vulnerability has been fixed in MLflow version 3.15.0. Security teams are being urged to upgrade immediately and review server logs, cloud metadata access and credential activity for signs of compromise, particularly where MLflow deployments are publicly reachable.