The Cybersecurity and Infrastructure Security Agency (CISA) has issued a warning to federal agencies about the active exploitation of a critical vulnerability in MLflow, an open-source AI engineering platform. This platform, which is backed by the Linux Foundation and widely used by organizations to manage AI applications, is currently being targeted by threat actors. The vulnerability, identified as CVE-2026-64849, involves a DNS-rebinding server-side request forgery bypass in MLflow's webhook delivery system. This flaw was addressed in the platform's version 3.15.0, yet unpatched instances remain vulnerable to unauthorized access to internal services and cloud metadata.
The MLflow security team highlighted that the default configuration of the MLflow Tracking Server exposes a webhooks API without authentication. This exposure allows attackers to send HTTP requests to sensitive endpoints and potentially access cloud instance metadata, including AWS IAM credentials, without complexity. In response to this threat, CISA has mandated U.S. Federal Civilian Executive Branch agencies to secure their MLflow instances within two weeks, following the directives of Binding Operational Directive 26-04.
This directive emphasizes the need for immediate action when vulnerabilities are publicly exposed, included in CISA's Known Exploited Vulnerabilities catalog, or capable of automated exploitation. Although further details from CISA are pending, cybersecurity firm watchTowr has noted that attackers started scanning for vulnerable MLflow systems almost immediately after the CVE was published. The firm advises organizations using MLflow to prioritize patching, review audit logs for potential breaches, and ensure sensitive credentials have not been exposed.

