MLflow is an open-source AI engineering platform widely utilized by data scientists, developers, and researchers in the field of machine learning and artificial intelligence. It is primarily used for managing the machine learning lifecycle, including experimentation, reproducibility, and deployment. Its utility in automating machine learning workflows has made it indispensable in various industry and academic settings. MLflow provides integration options with numerous tools and services, thereby facilitating a seamless AI model development process. Organizations relying on MLflow benefit from streamlined collaboration and enhanced productivity across teams. Despite its widespread usage, the security of this platform is paramount given its integration into sensitive data environments.
This scanner specifically targets the Server-Side Request Forgery (SSRF) vulnerability present in versions of MLflow prior to 3.17.0. SSRF vulnerabilities can be exploited by attackers to coerce a server into making unauthorized requests to internal or external domains. Such vulnerabilities often result from improper validation of user input that is used in crafting network requests. An attacker can manipulate this flaw to access restricted resources, potentially leading to sensitive data exposure or unauthorized actions. Through this scanner, users can detect the presence of this vulnerability and take necessary actions to protect their systems.
The vulnerability allows unauthorized third-party data deserialization and execution through invocation endpoints within MLflow. During the exploitation process, attackers utilize the `/ajax-api/3.0/mlflow/scorer/invoke` endpoint with maliciously crafted payloads. These payloads influence the MLflow server to interact with attacker-specified domains, potentially leaking internal service data. The exploitation relies on certain packaged modules being available, which further enables the injection of external requests using arbitrary headers and body content. This creates opportunities for attackers to probe internal networks indirectly and escalate their attack vectors depending on the exposed systems.
Exploitation of this SSRF vulnerability can have severe repercussions including unauthorized access to internal services and metadata endpoints within a cloud environment. Potential impacts also involve exposure of sensitive information and the ability for attackers to perform further pivoting within a network. The exploitation could allow attackers to map internal infrastructure, collect intelligence or even execute subsequent attacks using discovered endpoints. Such breaches may lead to data exfiltration, loss of system integrity, and unauthorized manipulation of network-accessible resources if unaddressed.
REFERENCES
- https://github.com/mlflow/mlflow/security/advisories/GHSA-2wqp-q3c6-ccj5
- https://github.com/mlflow/mlflow/commit/bac30b4148c4f40045b4d4db07ae60ab3838803f
- https://github.com/mlflow/mlflow/pull/25890
- https://github.com/mlflow/mlflow/releases/tag/v3.17.0
- https://nvd.nist.gov/vuln/detail/CVE-2026-115886
- Upgrade MLflow to version 3.17.0 or later to address the SSRF vulnerability by constraining third-party scorer classes and their constructor arguments.
- Implement network-level access controls to restrict outbound requests from MLflow servers to trusted hosts only.
- Review and audit the use of third-party packages in your MLflow environment and remove any unnecessary components that could introduce vulnerabilities.
- Continuously monitor network traffic from MLflow deployments for unusual patterns that might indicate exploitation attempts.
- Ensure that input validation is robust to prevent the injection of malicious parameters into server requests.
Get AI-powered remediation steps tailored to your asset.
Try AI Solutions →