A recent vulnerability in MLflow, an open-source AI engineering platform, has been exploited by threat actors to steal sensitive information, including credentials and secrets. The vulnerability, tracked as CVE-2026-64849 with a CVSS score of 9.3, involves an unauthenticated server-side request forgery (SSRF) issue that allows attackers to send HTTP requests to internal endpoints.
This defect exists because the default MLflow Tracking Server exposes its APIs without authentication, making it susceptible to exploitation. Since in-the-wild exploitation began shortly after the vulnerability was assigned, organizations running MLflow prior to version 3.15.0 are urged to patch affected systems and review audit logs for potential compromises.
The US cybersecurity agency CISA has added this vulnerability to its Known Exploited Vulnerabilities catalog, recommending that federal agencies address it within two weeks.